src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_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     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          vfitab;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm256_set1_pd(fr->ic->k_rf);
 129     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 130     crf              = _mm256_set1_pd(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[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     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 143     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm256_setzero_pd();
 179         fiy0             = _mm256_setzero_pd();
 180         fiz0             = _mm256_setzero_pd();
 181         fix1             = _mm256_setzero_pd();
 182         fiy1             = _mm256_setzero_pd();
 183         fiz1             = _mm256_setzero_pd();
 184         fix2             = _mm256_setzero_pd();
 185         fiy2             = _mm256_setzero_pd();
 186         fiz2             = _mm256_setzero_pd();
 187         fix3             = _mm256_setzero_pd();
 188         fiy3             = _mm256_setzero_pd();
 189         fiz3             = _mm256_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_pd();
 193         vvdwsum          = _mm256_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm256_sub_pd(ix0,jx0);
 216             dy00             = _mm256_sub_pd(iy0,jy0);
 217             dz00             = _mm256_sub_pd(iz0,jz0);
 218             dx10             = _mm256_sub_pd(ix1,jx0);
 219             dy10             = _mm256_sub_pd(iy1,jy0);
 220             dz10             = _mm256_sub_pd(iz1,jz0);
 221             dx20             = _mm256_sub_pd(ix2,jx0);
 222             dy20             = _mm256_sub_pd(iy2,jy0);
 223             dz20             = _mm256_sub_pd(iz2,jz0);
 224             dx30             = _mm256_sub_pd(ix3,jx0);
 225             dy30             = _mm256_sub_pd(iy3,jy0);
 226             dz30             = _mm256_sub_pd(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 230             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 231             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 232             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 233
 234             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 235             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 236             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 237             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 238
 239             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 240             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 241             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 245                                                                  charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm256_setzero_pd();
 252             fjy0             = _mm256_setzero_pd();
 253             fjz0             = _mm256_setzero_pd();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm256_mul_pd(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 263                                             vdwioffsetptr0+vdwjidx0B,
 264                                             vdwioffsetptr0+vdwjidx0C,
 265                                             vdwioffsetptr0+vdwjidx0D,
 266                                             &c6_00,&c12_00);
 267
 268             /* Calculate table index by multiplying r with table scale and truncate to integer */
 269             rt               = _mm256_mul_pd(r00,vftabscale);
 270             vfitab           = _mm256_cvttpd_epi32(rt);
 271             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 272             vfitab           = _mm_slli_epi32(vfitab,3);
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 277             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 278             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 279             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 280             Heps             = _mm256_mul_pd(vfeps,H);
 281             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 282             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 283             vvdw6            = _mm256_mul_pd(c6_00,VV);
 284             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 285             fvdw6            = _mm256_mul_pd(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 291             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 292             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 293             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 294             Heps             = _mm256_mul_pd(vfeps,H);
 295             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 296             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 297             vvdw12           = _mm256_mul_pd(c12_00,VV);
 298             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 299             fvdw12           = _mm256_mul_pd(c12_00,FF);
 300             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 301             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_pd(fscal,dx00);
 310             ty               = _mm256_mul_pd(fscal,dy00);
 311             tz               = _mm256_mul_pd(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm256_add_pd(fix0,tx);
 315             fiy0             = _mm256_add_pd(fiy0,ty);
 316             fiz0             = _mm256_add_pd(fiz0,tz);
 317
 318             fjx0             = _mm256_add_pd(fjx0,tx);
 319             fjy0             = _mm256_add_pd(fjy0,ty);
 320             fjz0             = _mm256_add_pd(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm256_mul_pd(iq1,jq0);
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 331             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm256_add_pd(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_pd(fscal,dx10);
 340             ty               = _mm256_mul_pd(fscal,dy10);
 341             tz               = _mm256_mul_pd(fscal,dz10);
 342
 343             /* Update vectorial force */
 344             fix1             = _mm256_add_pd(fix1,tx);
 345             fiy1             = _mm256_add_pd(fiy1,ty);
 346             fiz1             = _mm256_add_pd(fiz1,tz);
 347
 348             fjx0             = _mm256_add_pd(fjx0,tx);
 349             fjy0             = _mm256_add_pd(fjy0,ty);
 350             fjz0             = _mm256_add_pd(fjz0,tz);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm256_mul_pd(iq2,jq0);
 358
 359             /* REACTION-FIELD ELECTROSTATICS */
 360             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 361             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm256_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm256_mul_pd(fscal,dx20);
 370             ty               = _mm256_mul_pd(fscal,dy20);
 371             tz               = _mm256_mul_pd(fscal,dz20);
 372
 373             /* Update vectorial force */
 374             fix2             = _mm256_add_pd(fix2,tx);
 375             fiy2             = _mm256_add_pd(fiy2,ty);
 376             fiz2             = _mm256_add_pd(fiz2,tz);
 377
 378             fjx0             = _mm256_add_pd(fjx0,tx);
 379             fjy0             = _mm256_add_pd(fjy0,ty);
 380             fjz0             = _mm256_add_pd(fjz0,tz);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq30             = _mm256_mul_pd(iq3,jq0);
 388
 389             /* REACTION-FIELD ELECTROSTATICS */
 390             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 391             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm256_add_pd(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm256_mul_pd(fscal,dx30);
 400             ty               = _mm256_mul_pd(fscal,dy30);
 401             tz               = _mm256_mul_pd(fscal,dz30);
 402
 403             /* Update vectorial force */
 404             fix3             = _mm256_add_pd(fix3,tx);
 405             fiy3             = _mm256_add_pd(fiy3,ty);
 406             fiz3             = _mm256_add_pd(fiz3,tz);
 407
 408             fjx0             = _mm256_add_pd(fjx0,tx);
 409             fjy0             = _mm256_add_pd(fjy0,ty);
 410             fjz0             = _mm256_add_pd(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416
 417             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 155 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             /* Sign of each element will be negative for non-real atoms.
 431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 432              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 433              */
 434             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 435
 436             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 437             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 438             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 439
 440             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 441             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 442             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 443             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 444             j_coord_offsetA  = DIM*jnrA;
 445             j_coord_offsetB  = DIM*jnrB;
 446             j_coord_offsetC  = DIM*jnrC;
 447             j_coord_offsetD  = DIM*jnrD;
 448
 449             /* load j atom coordinates */
 450             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 451                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 452                                                  &jx0,&jy0,&jz0);
 453
 454             /* Calculate displacement vector */
 455             dx00             = _mm256_sub_pd(ix0,jx0);
 456             dy00             = _mm256_sub_pd(iy0,jy0);
 457             dz00             = _mm256_sub_pd(iz0,jz0);
 458             dx10             = _mm256_sub_pd(ix1,jx0);
 459             dy10             = _mm256_sub_pd(iy1,jy0);
 460             dz10             = _mm256_sub_pd(iz1,jz0);
 461             dx20             = _mm256_sub_pd(ix2,jx0);
 462             dy20             = _mm256_sub_pd(iy2,jy0);
 463             dz20             = _mm256_sub_pd(iz2,jz0);
 464             dx30             = _mm256_sub_pd(ix3,jx0);
 465             dy30             = _mm256_sub_pd(iy3,jy0);
 466             dz30             = _mm256_sub_pd(iz3,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 470             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 471             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 472             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 473
 474             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 475             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 476             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 477             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 478
 479             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 480             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 481             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 485                                                                  charge+jnrC+0,charge+jnrD+0);
 486             vdwjidx0A        = 2*vdwtype[jnrA+0];
 487             vdwjidx0B        = 2*vdwtype[jnrB+0];
 488             vdwjidx0C        = 2*vdwtype[jnrC+0];
 489             vdwjidx0D        = 2*vdwtype[jnrD+0];
 490
 491             fjx0             = _mm256_setzero_pd();
 492             fjy0             = _mm256_setzero_pd();
 493             fjz0             = _mm256_setzero_pd();
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r00              = _mm256_mul_pd(rsq00,rinv00);
 500             r00              = _mm256_andnot_pd(dummy_mask,r00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 504                                             vdwioffsetptr0+vdwjidx0B,
 505                                             vdwioffsetptr0+vdwjidx0C,
 506                                             vdwioffsetptr0+vdwjidx0D,
 507                                             &c6_00,&c12_00);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm256_mul_pd(r00,vftabscale);
 511             vfitab           = _mm256_cvttpd_epi32(rt);
 512             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 513             vfitab           = _mm_slli_epi32(vfitab,3);
 514
 515             /* CUBIC SPLINE TABLE DISPERSION */
 516             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 517             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 518             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 519             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 520             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 521             Heps             = _mm256_mul_pd(vfeps,H);
 522             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 523             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 524             vvdw6            = _mm256_mul_pd(c6_00,VV);
 525             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 526             fvdw6            = _mm256_mul_pd(c6_00,FF);
 527
 528             /* CUBIC SPLINE TABLE REPULSION */
 529             vfitab           = _mm_add_epi32(vfitab,ifour);
 530             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 531             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 532             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 533             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 534             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 535             Heps             = _mm256_mul_pd(vfeps,H);
 536             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 537             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 538             vvdw12           = _mm256_mul_pd(c12_00,VV);
 539             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 540             fvdw12           = _mm256_mul_pd(c12_00,FF);
 541             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 542             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 546             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 547
 548             fscal            = fvdw;
 549
 550             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx00);
 554             ty               = _mm256_mul_pd(fscal,dy00);
 555             tz               = _mm256_mul_pd(fscal,dz00);
 556
 557             /* Update vectorial force */
 558             fix0             = _mm256_add_pd(fix0,tx);
 559             fiy0             = _mm256_add_pd(fiy0,ty);
 560             fiz0             = _mm256_add_pd(fiz0,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             qq10             = _mm256_mul_pd(iq1,jq0);
 572
 573             /* REACTION-FIELD ELECTROSTATICS */
 574             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 575             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),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,dx10);
 587             ty               = _mm256_mul_pd(fscal,dy10);
 588             tz               = _mm256_mul_pd(fscal,dz10);
 589
 590             /* Update vectorial force */
 591             fix1             = _mm256_add_pd(fix1,tx);
 592             fiy1             = _mm256_add_pd(fiy1,ty);
 593             fiz1             = _mm256_add_pd(fiz1,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             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq20             = _mm256_mul_pd(iq2,jq0);
 605
 606             /* REACTION-FIELD ELECTROSTATICS */
 607             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 608             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velec            = _mm256_andnot_pd(dummy_mask,velec);
 612             velecsum         = _mm256_add_pd(velecsum,velec);
 613
 614             fscal            = felec;
 615
 616             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm256_mul_pd(fscal,dx20);
 620             ty               = _mm256_mul_pd(fscal,dy20);
 621             tz               = _mm256_mul_pd(fscal,dz20);
 622
 623             /* Update vectorial force */
 624             fix2             = _mm256_add_pd(fix2,tx);
 625             fiy2             = _mm256_add_pd(fiy2,ty);
 626             fiz2             = _mm256_add_pd(fiz2,tz);
 627
 628             fjx0             = _mm256_add_pd(fjx0,tx);
 629             fjy0             = _mm256_add_pd(fjy0,ty);
 630             fjz0             = _mm256_add_pd(fjz0,tz);
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq30             = _mm256_mul_pd(iq3,jq0);
 638
 639             /* REACTION-FIELD ELECTROSTATICS */
 640             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 641             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm256_andnot_pd(dummy_mask,velec);
 645             velecsum         = _mm256_add_pd(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm256_mul_pd(fscal,dx30);
 653             ty               = _mm256_mul_pd(fscal,dy30);
 654             tz               = _mm256_mul_pd(fscal,dz30);
 655
 656             /* Update vectorial force */
 657             fix3             = _mm256_add_pd(fix3,tx);
 658             fiy3             = _mm256_add_pd(fiy3,ty);
 659             fiz3             = _mm256_add_pd(fiz3,tz);
 660
 661             fjx0             = _mm256_add_pd(fjx0,tx);
 662             fjy0             = _mm256_add_pd(fjy0,ty);
 663             fjz0             = _mm256_add_pd(fjz0,tz);
 664
 665             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 666             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 667             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 668             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 669
 670             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 671
 672             /* Inner loop uses 156 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 678                                                  f+i_coord_offset,fshift+i_shift_offset);
 679
 680         ggid                        = gid[iidx];
 681         /* Update potential energies */
 682         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 683         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 26 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
 697 }
 698 /*
 699  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
 700  * Electrostatics interaction: ReactionField
 701  * VdW interaction:            CubicSplineTable
 702  * Geometry:                   Water4-Particle
 703  * Calculate force/pot:        Force
 704  */
 705 void
 706 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
 707                     (t_nblist                    * gmx_restrict       nlist,
 708                      rvec                        * gmx_restrict          xx,
 709                      rvec                        * gmx_restrict          ff,
 710                      t_forcerec                  * gmx_restrict          fr,
 711                      t_mdatoms                   * gmx_restrict     mdatoms,
 712                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 713                      t_nrnb                      * gmx_restrict        nrnb)
 714 {
 715     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 716      * just 0 for non-waters.
 717      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 718      * jnr indices corresponding to data put in the four positions in the SIMD register.
 719      */
 720     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 721     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 722     int              jnrA,jnrB,jnrC,jnrD;
 723     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 724     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 725     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 726     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 727     real             rcutoff_scalar;
 728     real             *shiftvec,*fshift,*x,*f;
 729     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 730     real             scratch[4*DIM];
 731     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 732     real *           vdwioffsetptr0;
 733     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 734     real *           vdwioffsetptr1;
 735     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 736     real *           vdwioffsetptr2;
 737     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 738     real *           vdwioffsetptr3;
 739     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 740     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 741     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 742     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 743     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 744     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 745     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 746     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 747     real             *charge;
 748     int              nvdwtype;
 749     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 750     int              *vdwtype;
 751     real             *vdwparam;
 752     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 753     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 754     __m128i          vfitab;
 755     __m128i          ifour       = _mm_set1_epi32(4);
 756     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 757     real             *vftab;
 758     __m256d          dummy_mask,cutoff_mask;
 759     __m128           tmpmask0,tmpmask1;
 760     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 761     __m256d          one     = _mm256_set1_pd(1.0);
 762     __m256d          two     = _mm256_set1_pd(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm256_set1_pd(fr->epsfac);
 775     charge           = mdatoms->chargeA;
 776     krf              = _mm256_set1_pd(fr->ic->k_rf);
 777     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 778     crf              = _mm256_set1_pd(fr->ic->c_rf);
 779     nvdwtype         = fr->ntype;
 780     vdwparam         = fr->nbfp;
 781     vdwtype          = mdatoms->typeA;
 782
 783     vftab            = kernel_data->table_vdw->data;
 784     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 789     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 790     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 791     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 792
 793     /* Avoid stupid compiler warnings */
 794     jnrA = jnrB = jnrC = jnrD = 0;
 795     j_coord_offsetA = 0;
 796     j_coord_offsetB = 0;
 797     j_coord_offsetC = 0;
 798     j_coord_offsetD = 0;
 799
 800     outeriter        = 0;
 801     inneriter        = 0;
 802
 803     for(iidx=0;iidx<4*DIM;iidx++)
 804     {
 805         scratch[iidx] = 0.0;
 806     }
 807
 808     /* Start outer loop over neighborlists */
 809     for(iidx=0; iidx<nri; iidx++)
 810     {
 811         /* Load shift vector for this list */
 812         i_shift_offset   = DIM*shiftidx[iidx];
 813
 814         /* Load limits for loop over neighbors */
 815         j_index_start    = jindex[iidx];
 816         j_index_end      = jindex[iidx+1];
 817
 818         /* Get outer coordinate index */
 819         inr              = iinr[iidx];
 820         i_coord_offset   = DIM*inr;
 821
 822         /* Load i particle coords and add shift vector */
 823         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 824                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 825
 826         fix0             = _mm256_setzero_pd();
 827         fiy0             = _mm256_setzero_pd();
 828         fiz0             = _mm256_setzero_pd();
 829         fix1             = _mm256_setzero_pd();
 830         fiy1             = _mm256_setzero_pd();
 831         fiz1             = _mm256_setzero_pd();
 832         fix2             = _mm256_setzero_pd();
 833         fiy2             = _mm256_setzero_pd();
 834         fiz2             = _mm256_setzero_pd();
 835         fix3             = _mm256_setzero_pd();
 836         fiy3             = _mm256_setzero_pd();
 837         fiz3             = _mm256_setzero_pd();
 838
 839         /* Start inner kernel loop */
 840         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrA             = jjnr[jidx];
 845             jnrB             = jjnr[jidx+1];
 846             jnrC             = jjnr[jidx+2];
 847             jnrD             = jjnr[jidx+3];
 848             j_coord_offsetA  = DIM*jnrA;
 849             j_coord_offsetB  = DIM*jnrB;
 850             j_coord_offsetC  = DIM*jnrC;
 851             j_coord_offsetD  = DIM*jnrD;
 852
 853             /* load j atom coordinates */
 854             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 855                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 856                                                  &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx00             = _mm256_sub_pd(ix0,jx0);
 860             dy00             = _mm256_sub_pd(iy0,jy0);
 861             dz00             = _mm256_sub_pd(iz0,jz0);
 862             dx10             = _mm256_sub_pd(ix1,jx0);
 863             dy10             = _mm256_sub_pd(iy1,jy0);
 864             dz10             = _mm256_sub_pd(iz1,jz0);
 865             dx20             = _mm256_sub_pd(ix2,jx0);
 866             dy20             = _mm256_sub_pd(iy2,jy0);
 867             dz20             = _mm256_sub_pd(iz2,jz0);
 868             dx30             = _mm256_sub_pd(ix3,jx0);
 869             dy30             = _mm256_sub_pd(iy3,jy0);
 870             dz30             = _mm256_sub_pd(iz3,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 874             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 875             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 876             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 877
 878             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 879             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 880             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 881             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 882
 883             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 884             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 885             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 886
 887             /* Load parameters for j particles */
 888             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 889                                                                  charge+jnrC+0,charge+jnrD+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892             vdwjidx0C        = 2*vdwtype[jnrC+0];
 893             vdwjidx0D        = 2*vdwtype[jnrD+0];
 894
 895             fjx0             = _mm256_setzero_pd();
 896             fjy0             = _mm256_setzero_pd();
 897             fjz0             = _mm256_setzero_pd();
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r00              = _mm256_mul_pd(rsq00,rinv00);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 907                                             vdwioffsetptr0+vdwjidx0B,
 908                                             vdwioffsetptr0+vdwjidx0C,
 909                                             vdwioffsetptr0+vdwjidx0D,
 910                                             &c6_00,&c12_00);
 911
 912             /* Calculate table index by multiplying r with table scale and truncate to integer */
 913             rt               = _mm256_mul_pd(r00,vftabscale);
 914             vfitab           = _mm256_cvttpd_epi32(rt);
 915             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 916             vfitab           = _mm_slli_epi32(vfitab,3);
 917
 918             /* CUBIC SPLINE TABLE DISPERSION */
 919             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 920             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 921             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 922             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 923             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 924             Heps             = _mm256_mul_pd(vfeps,H);
 925             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 926             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 927             fvdw6            = _mm256_mul_pd(c6_00,FF);
 928
 929             /* CUBIC SPLINE TABLE REPULSION */
 930             vfitab           = _mm_add_epi32(vfitab,ifour);
 931             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 932             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 933             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 934             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 935             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 936             Heps             = _mm256_mul_pd(vfeps,H);
 937             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 938             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 939             fvdw12           = _mm256_mul_pd(c12_00,FF);
 940             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 941
 942             fscal            = fvdw;
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm256_mul_pd(fscal,dx00);
 946             ty               = _mm256_mul_pd(fscal,dy00);
 947             tz               = _mm256_mul_pd(fscal,dz00);
 948
 949             /* Update vectorial force */
 950             fix0             = _mm256_add_pd(fix0,tx);
 951             fiy0             = _mm256_add_pd(fiy0,ty);
 952             fiz0             = _mm256_add_pd(fiz0,tz);
 953
 954             fjx0             = _mm256_add_pd(fjx0,tx);
 955             fjy0             = _mm256_add_pd(fjy0,ty);
 956             fjz0             = _mm256_add_pd(fjz0,tz);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq10             = _mm256_mul_pd(iq1,jq0);
 964
 965             /* REACTION-FIELD ELECTROSTATICS */
 966             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 967
 968             fscal            = felec;
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm256_mul_pd(fscal,dx10);
 972             ty               = _mm256_mul_pd(fscal,dy10);
 973             tz               = _mm256_mul_pd(fscal,dz10);
 974
 975             /* Update vectorial force */
 976             fix1             = _mm256_add_pd(fix1,tx);
 977             fiy1             = _mm256_add_pd(fiy1,ty);
 978             fiz1             = _mm256_add_pd(fiz1,tz);
 979
 980             fjx0             = _mm256_add_pd(fjx0,tx);
 981             fjy0             = _mm256_add_pd(fjy0,ty);
 982             fjz0             = _mm256_add_pd(fjz0,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq20             = _mm256_mul_pd(iq2,jq0);
 990
 991             /* REACTION-FIELD ELECTROSTATICS */
 992             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm256_mul_pd(fscal,dx20);
 998             ty               = _mm256_mul_pd(fscal,dy20);
 999             tz               = _mm256_mul_pd(fscal,dz20);
1000
1001             /* Update vectorial force */
1002             fix2             = _mm256_add_pd(fix2,tx);
1003             fiy2             = _mm256_add_pd(fiy2,ty);
1004             fiz2             = _mm256_add_pd(fiz2,tz);
1005
1006             fjx0             = _mm256_add_pd(fjx0,tx);
1007             fjy0             = _mm256_add_pd(fjy0,ty);
1008             fjz0             = _mm256_add_pd(fjz0,tz);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq30             = _mm256_mul_pd(iq3,jq0);
1016
1017             /* REACTION-FIELD ELECTROSTATICS */
1018             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1019
1020             fscal            = felec;
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm256_mul_pd(fscal,dx30);
1024             ty               = _mm256_mul_pd(fscal,dy30);
1025             tz               = _mm256_mul_pd(fscal,dz30);
1026
1027             /* Update vectorial force */
1028             fix3             = _mm256_add_pd(fix3,tx);
1029             fiy3             = _mm256_add_pd(fiy3,ty);
1030             fiz3             = _mm256_add_pd(fiz3,tz);
1031
1032             fjx0             = _mm256_add_pd(fjx0,tx);
1033             fjy0             = _mm256_add_pd(fjy0,ty);
1034             fjz0             = _mm256_add_pd(fjz0,tz);
1035
1036             fjptrA             = f+j_coord_offsetA;
1037             fjptrB             = f+j_coord_offsetB;
1038             fjptrC             = f+j_coord_offsetC;
1039             fjptrD             = f+j_coord_offsetD;
1040
1041             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1042
1043             /* Inner loop uses 132 flops */
1044         }
1045
1046         if(jidx<j_index_end)
1047         {
1048
1049             /* Get j neighbor index, and coordinate index */
1050             jnrlistA         = jjnr[jidx];
1051             jnrlistB         = jjnr[jidx+1];
1052             jnrlistC         = jjnr[jidx+2];
1053             jnrlistD         = jjnr[jidx+3];
1054             /* Sign of each element will be negative for non-real atoms.
1055              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1056              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1057              */
1058             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1059
1060             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1061             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1062             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1063
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070             j_coord_offsetC  = DIM*jnrC;
1071             j_coord_offsetD  = DIM*jnrD;
1072
1073             /* load j atom coordinates */
1074             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1075                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1076                                                  &jx0,&jy0,&jz0);
1077
1078             /* Calculate displacement vector */
1079             dx00             = _mm256_sub_pd(ix0,jx0);
1080             dy00             = _mm256_sub_pd(iy0,jy0);
1081             dz00             = _mm256_sub_pd(iz0,jz0);
1082             dx10             = _mm256_sub_pd(ix1,jx0);
1083             dy10             = _mm256_sub_pd(iy1,jy0);
1084             dz10             = _mm256_sub_pd(iz1,jz0);
1085             dx20             = _mm256_sub_pd(ix2,jx0);
1086             dy20             = _mm256_sub_pd(iy2,jy0);
1087             dz20             = _mm256_sub_pd(iz2,jz0);
1088             dx30             = _mm256_sub_pd(ix3,jx0);
1089             dy30             = _mm256_sub_pd(iy3,jy0);
1090             dz30             = _mm256_sub_pd(iz3,jz0);
1091
1092             /* Calculate squared distance and things based on it */
1093             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1094             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1095             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1096             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1097
1098             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1099             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1100             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1101             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1102
1103             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1104             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1105             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1106
1107             /* Load parameters for j particles */
1108             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1109                                                                  charge+jnrC+0,charge+jnrD+0);
1110             vdwjidx0A        = 2*vdwtype[jnrA+0];
1111             vdwjidx0B        = 2*vdwtype[jnrB+0];
1112             vdwjidx0C        = 2*vdwtype[jnrC+0];
1113             vdwjidx0D        = 2*vdwtype[jnrD+0];
1114
1115             fjx0             = _mm256_setzero_pd();
1116             fjy0             = _mm256_setzero_pd();
1117             fjz0             = _mm256_setzero_pd();
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             r00              = _mm256_mul_pd(rsq00,rinv00);
1124             r00              = _mm256_andnot_pd(dummy_mask,r00);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1128                                             vdwioffsetptr0+vdwjidx0B,
1129                                             vdwioffsetptr0+vdwjidx0C,
1130                                             vdwioffsetptr0+vdwjidx0D,
1131                                             &c6_00,&c12_00);
1132
1133             /* Calculate table index by multiplying r with table scale and truncate to integer */
1134             rt               = _mm256_mul_pd(r00,vftabscale);
1135             vfitab           = _mm256_cvttpd_epi32(rt);
1136             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1137             vfitab           = _mm_slli_epi32(vfitab,3);
1138
1139             /* CUBIC SPLINE TABLE DISPERSION */
1140             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1141             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1142             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1143             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1144             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1145             Heps             = _mm256_mul_pd(vfeps,H);
1146             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1147             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1148             fvdw6            = _mm256_mul_pd(c6_00,FF);
1149
1150             /* CUBIC SPLINE TABLE REPULSION */
1151             vfitab           = _mm_add_epi32(vfitab,ifour);
1152             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1153             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1154             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1155             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1156             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1157             Heps             = _mm256_mul_pd(vfeps,H);
1158             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1159             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1160             fvdw12           = _mm256_mul_pd(c12_00,FF);
1161             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1162
1163             fscal            = fvdw;
1164
1165             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm256_mul_pd(fscal,dx00);
1169             ty               = _mm256_mul_pd(fscal,dy00);
1170             tz               = _mm256_mul_pd(fscal,dz00);
1171
1172             /* Update vectorial force */
1173             fix0             = _mm256_add_pd(fix0,tx);
1174             fiy0             = _mm256_add_pd(fiy0,ty);
1175             fiz0             = _mm256_add_pd(fiz0,tz);
1176
1177             fjx0             = _mm256_add_pd(fjx0,tx);
1178             fjy0             = _mm256_add_pd(fjy0,ty);
1179             fjz0             = _mm256_add_pd(fjz0,tz);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq10             = _mm256_mul_pd(iq1,jq0);
1187
1188             /* REACTION-FIELD ELECTROSTATICS */
1189             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1190
1191             fscal            = felec;
1192
1193             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm256_mul_pd(fscal,dx10);
1197             ty               = _mm256_mul_pd(fscal,dy10);
1198             tz               = _mm256_mul_pd(fscal,dz10);
1199
1200             /* Update vectorial force */
1201             fix1             = _mm256_add_pd(fix1,tx);
1202             fiy1             = _mm256_add_pd(fiy1,ty);
1203             fiz1             = _mm256_add_pd(fiz1,tz);
1204
1205             fjx0             = _mm256_add_pd(fjx0,tx);
1206             fjy0             = _mm256_add_pd(fjy0,ty);
1207             fjz0             = _mm256_add_pd(fjz0,tz);
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             /* Compute parameters for interactions between i and j atoms */
1214             qq20             = _mm256_mul_pd(iq2,jq0);
1215
1216             /* REACTION-FIELD ELECTROSTATICS */
1217             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1218
1219             fscal            = felec;
1220
1221             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm256_mul_pd(fscal,dx20);
1225             ty               = _mm256_mul_pd(fscal,dy20);
1226             tz               = _mm256_mul_pd(fscal,dz20);
1227
1228             /* Update vectorial force */
1229             fix2             = _mm256_add_pd(fix2,tx);
1230             fiy2             = _mm256_add_pd(fiy2,ty);
1231             fiz2             = _mm256_add_pd(fiz2,tz);
1232
1233             fjx0             = _mm256_add_pd(fjx0,tx);
1234             fjy0             = _mm256_add_pd(fjy0,ty);
1235             fjz0             = _mm256_add_pd(fjz0,tz);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             /* Compute parameters for interactions between i and j atoms */
1242             qq30             = _mm256_mul_pd(iq3,jq0);
1243
1244             /* REACTION-FIELD ELECTROSTATICS */
1245             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1246
1247             fscal            = felec;
1248
1249             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1250
1251             /* Calculate temporary vectorial force */
1252             tx               = _mm256_mul_pd(fscal,dx30);
1253             ty               = _mm256_mul_pd(fscal,dy30);
1254             tz               = _mm256_mul_pd(fscal,dz30);
1255
1256             /* Update vectorial force */
1257             fix3             = _mm256_add_pd(fix3,tx);
1258             fiy3             = _mm256_add_pd(fiy3,ty);
1259             fiz3             = _mm256_add_pd(fiz3,tz);
1260
1261             fjx0             = _mm256_add_pd(fjx0,tx);
1262             fjy0             = _mm256_add_pd(fjy0,ty);
1263             fjz0             = _mm256_add_pd(fjz0,tz);
1264
1265             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1266             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1267             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1268             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1269
1270             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1271
1272             /* Inner loop uses 133 flops */
1273         }
1274
1275         /* End of innermost loop */
1276
1277         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1278                                                  f+i_coord_offset,fshift+i_shift_offset);
1279
1280         /* Increment number of inner iterations */
1281         inneriter                  += j_index_end - j_index_start;
1282
1283         /* Outer loop uses 24 flops */
1284     }
1285
1286     /* Increment number of outer iterations */
1287     outeriter        += nri;
1288
1289     /* Update outer/inner flops */
1290
1291     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);
1292 }