src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 145     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_pd();
 181         fiy0             = _mm256_setzero_pd();
 182         fiz0             = _mm256_setzero_pd();
 183         fix1             = _mm256_setzero_pd();
 184         fiy1             = _mm256_setzero_pd();
 185         fiz1             = _mm256_setzero_pd();
 186         fix2             = _mm256_setzero_pd();
 187         fiy2             = _mm256_setzero_pd();
 188         fiz2             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217             dx10             = _mm256_sub_pd(ix1,jx0);
 218             dy10             = _mm256_sub_pd(iy1,jy0);
 219             dz10             = _mm256_sub_pd(iz1,jz0);
 220             dx20             = _mm256_sub_pd(ix2,jx0);
 221             dy20             = _mm256_sub_pd(iy2,jy0);
 222             dz20             = _mm256_sub_pd(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 232
 233             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 234             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 235             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 239                                                                  charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm256_setzero_pd();
 246             fjy0             = _mm256_setzero_pd();
 247             fjz0             = _mm256_setzero_pd();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 254             {
 255
 256             r00              = _mm256_mul_pd(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm256_mul_pd(iq0,jq0);
 260             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 261                                             vdwioffsetptr0+vdwjidx0B,
 262                                             vdwioffsetptr0+vdwjidx0C,
 263                                             vdwioffsetptr0+vdwjidx0D,
 264                                             &c6_00,&c12_00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm256_mul_pd(r00,vftabscale);
 268             vfitab           = _mm256_cvttpd_epi32(rt);
 269             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 270             vfitab           = _mm_slli_epi32(vfitab,3);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 274             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 278             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 279             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 280             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 281             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 282             Heps             = _mm256_mul_pd(vfeps,H);
 283             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 284             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 285             vvdw6            = _mm256_mul_pd(c6_00,VV);
 286             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 287             fvdw6            = _mm256_mul_pd(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 293             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 294             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 295             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 296             Heps             = _mm256_mul_pd(vfeps,H);
 297             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 298             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 299             vvdw12           = _mm256_mul_pd(c12_00,VV);
 300             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 301             fvdw12           = _mm256_mul_pd(c12_00,FF);
 302             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 303             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 304
 305             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm256_and_pd(velec,cutoff_mask);
 309             velecsum         = _mm256_add_pd(velecsum,velec);
 310             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 311             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 312
 313             fscal            = _mm256_add_pd(felec,fvdw);
 314
 315             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_pd(fscal,dx00);
 319             ty               = _mm256_mul_pd(fscal,dy00);
 320             tz               = _mm256_mul_pd(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm256_add_pd(fix0,tx);
 324             fiy0             = _mm256_add_pd(fiy0,ty);
 325             fiz0             = _mm256_add_pd(fiz0,tz);
 326
 327             fjx0             = _mm256_add_pd(fjx0,tx);
 328             fjy0             = _mm256_add_pd(fjy0,ty);
 329             fjz0             = _mm256_add_pd(fjz0,tz);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 338             {
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm256_mul_pd(iq1,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 345             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 346
 347             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm256_and_pd(velec,cutoff_mask);
 351             velecsum         = _mm256_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm256_mul_pd(fscal,dx10);
 359             ty               = _mm256_mul_pd(fscal,dy10);
 360             tz               = _mm256_mul_pd(fscal,dz10);
 361
 362             /* Update vectorial force */
 363             fix1             = _mm256_add_pd(fix1,tx);
 364             fiy1             = _mm256_add_pd(fiy1,ty);
 365             fiz1             = _mm256_add_pd(fiz1,tz);
 366
 367             fjx0             = _mm256_add_pd(fjx0,tx);
 368             fjy0             = _mm256_add_pd(fjy0,ty);
 369             fjz0             = _mm256_add_pd(fjz0,tz);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 378             {
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq20             = _mm256_mul_pd(iq2,jq0);
 382
 383             /* REACTION-FIELD ELECTROSTATICS */
 384             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 385             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 386
 387             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm256_and_pd(velec,cutoff_mask);
 391             velecsum         = _mm256_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_pd(fscal,dx20);
 399             ty               = _mm256_mul_pd(fscal,dy20);
 400             tz               = _mm256_mul_pd(fscal,dz20);
 401
 402             /* Update vectorial force */
 403             fix2             = _mm256_add_pd(fix2,tx);
 404             fiy2             = _mm256_add_pd(fiy2,ty);
 405             fiz2             = _mm256_add_pd(fiz2,tz);
 406
 407             fjx0             = _mm256_add_pd(fjx0,tx);
 408             fjy0             = _mm256_add_pd(fjy0,ty);
 409             fjz0             = _mm256_add_pd(fjz0,tz);
 410
 411             }
 412
 413             fjptrA             = f+j_coord_offsetA;
 414             fjptrB             = f+j_coord_offsetB;
 415             fjptrC             = f+j_coord_offsetC;
 416             fjptrD             = f+j_coord_offsetD;
 417
 418             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 419
 420             /* Inner loop uses 147 flops */
 421         }
 422
 423         if(jidx<j_index_end)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrlistA         = jjnr[jidx];
 428             jnrlistB         = jjnr[jidx+1];
 429             jnrlistC         = jjnr[jidx+2];
 430             jnrlistD         = jjnr[jidx+3];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 434              */
 435             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 436
 437             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 438             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 439             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 440
 441             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 442             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 443             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 444             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 445             j_coord_offsetA  = DIM*jnrA;
 446             j_coord_offsetB  = DIM*jnrB;
 447             j_coord_offsetC  = DIM*jnrC;
 448             j_coord_offsetD  = DIM*jnrD;
 449
 450             /* load j atom coordinates */
 451             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 452                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 453                                                  &jx0,&jy0,&jz0);
 454
 455             /* Calculate displacement vector */
 456             dx00             = _mm256_sub_pd(ix0,jx0);
 457             dy00             = _mm256_sub_pd(iy0,jy0);
 458             dz00             = _mm256_sub_pd(iz0,jz0);
 459             dx10             = _mm256_sub_pd(ix1,jx0);
 460             dy10             = _mm256_sub_pd(iy1,jy0);
 461             dz10             = _mm256_sub_pd(iz1,jz0);
 462             dx20             = _mm256_sub_pd(ix2,jx0);
 463             dy20             = _mm256_sub_pd(iy2,jy0);
 464             dz20             = _mm256_sub_pd(iz2,jz0);
 465
 466             /* Calculate squared distance and things based on it */
 467             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 468             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 469             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 470
 471             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 472             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 473             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 474
 475             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 476             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 477             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 478
 479             /* Load parameters for j particles */
 480             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 481                                                                  charge+jnrC+0,charge+jnrD+0);
 482             vdwjidx0A        = 2*vdwtype[jnrA+0];
 483             vdwjidx0B        = 2*vdwtype[jnrB+0];
 484             vdwjidx0C        = 2*vdwtype[jnrC+0];
 485             vdwjidx0D        = 2*vdwtype[jnrD+0];
 486
 487             fjx0             = _mm256_setzero_pd();
 488             fjy0             = _mm256_setzero_pd();
 489             fjz0             = _mm256_setzero_pd();
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 496             {
 497
 498             r00              = _mm256_mul_pd(rsq00,rinv00);
 499             r00              = _mm256_andnot_pd(dummy_mask,r00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq00             = _mm256_mul_pd(iq0,jq0);
 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             /* REACTION-FIELD ELECTROSTATICS */
 516             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 517             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 518
 519             /* CUBIC SPLINE TABLE DISPERSION */
 520             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 521             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 522             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 523             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 524             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 525             Heps             = _mm256_mul_pd(vfeps,H);
 526             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 527             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 528             vvdw6            = _mm256_mul_pd(c6_00,VV);
 529             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 530             fvdw6            = _mm256_mul_pd(c6_00,FF);
 531
 532             /* CUBIC SPLINE TABLE REPULSION */
 533             vfitab           = _mm_add_epi32(vfitab,ifour);
 534             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 535             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 536             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 537             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 538             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 539             Heps             = _mm256_mul_pd(vfeps,H);
 540             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 541             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 542             vvdw12           = _mm256_mul_pd(c12_00,VV);
 543             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 544             fvdw12           = _mm256_mul_pd(c12_00,FF);
 545             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 546             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 547
 548             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm256_and_pd(velec,cutoff_mask);
 552             velec            = _mm256_andnot_pd(dummy_mask,velec);
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 555             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 556             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 557
 558             fscal            = _mm256_add_pd(felec,fvdw);
 559
 560             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 561
 562             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm256_mul_pd(fscal,dx00);
 566             ty               = _mm256_mul_pd(fscal,dy00);
 567             tz               = _mm256_mul_pd(fscal,dz00);
 568
 569             /* Update vectorial force */
 570             fix0             = _mm256_add_pd(fix0,tx);
 571             fiy0             = _mm256_add_pd(fiy0,ty);
 572             fiz0             = _mm256_add_pd(fiz0,tz);
 573
 574             fjx0             = _mm256_add_pd(fjx0,tx);
 575             fjy0             = _mm256_add_pd(fjy0,ty);
 576             fjz0             = _mm256_add_pd(fjz0,tz);
 577
 578             }
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 585             {
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm256_mul_pd(iq1,jq0);
 589
 590             /* REACTION-FIELD ELECTROSTATICS */
 591             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 592             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 593
 594             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_and_pd(velec,cutoff_mask);
 598             velec            = _mm256_andnot_pd(dummy_mask,velec);
 599             velecsum         = _mm256_add_pd(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 604
 605             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm256_mul_pd(fscal,dx10);
 609             ty               = _mm256_mul_pd(fscal,dy10);
 610             tz               = _mm256_mul_pd(fscal,dz10);
 611
 612             /* Update vectorial force */
 613             fix1             = _mm256_add_pd(fix1,tx);
 614             fiy1             = _mm256_add_pd(fiy1,ty);
 615             fiz1             = _mm256_add_pd(fiz1,tz);
 616
 617             fjx0             = _mm256_add_pd(fjx0,tx);
 618             fjy0             = _mm256_add_pd(fjy0,ty);
 619             fjz0             = _mm256_add_pd(fjz0,tz);
 620
 621             }
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 628             {
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq20             = _mm256_mul_pd(iq2,jq0);
 632
 633             /* REACTION-FIELD ELECTROSTATICS */
 634             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 635             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 636
 637             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_and_pd(velec,cutoff_mask);
 641             velec            = _mm256_andnot_pd(dummy_mask,velec);
 642             velecsum         = _mm256_add_pd(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 647
 648             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_pd(fscal,dx20);
 652             ty               = _mm256_mul_pd(fscal,dy20);
 653             tz               = _mm256_mul_pd(fscal,dz20);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm256_add_pd(fix2,tx);
 657             fiy2             = _mm256_add_pd(fiy2,ty);
 658             fiz2             = _mm256_add_pd(fiz2,tz);
 659
 660             fjx0             = _mm256_add_pd(fjx0,tx);
 661             fjy0             = _mm256_add_pd(fjy0,ty);
 662             fjz0             = _mm256_add_pd(fjz0,tz);
 663
 664             }
 665
 666             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 667             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 668             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 669             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 670
 671             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 672
 673             /* Inner loop uses 148 flops */
 674         }
 675
 676         /* End of innermost loop */
 677
 678         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 679                                                  f+i_coord_offset,fshift+i_shift_offset);
 680
 681         ggid                        = gid[iidx];
 682         /* Update potential energies */
 683         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 684         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 685
 686         /* Increment number of inner iterations */
 687         inneriter                  += j_index_end - j_index_start;
 688
 689         /* Outer loop uses 20 flops */
 690     }
 691
 692     /* Increment number of outer iterations */
 693     outeriter        += nri;
 694
 695     /* Update outer/inner flops */
 696
 697     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 698 }
 699 /*
 700  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 701  * Electrostatics interaction: ReactionField
 702  * VdW interaction:            CubicSplineTable
 703  * Geometry:                   Water3-Particle
 704  * Calculate force/pot:        Force
 705  */
 706 void
 707 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 708                     (t_nblist                    * gmx_restrict       nlist,
 709                      rvec                        * gmx_restrict          xx,
 710                      rvec                        * gmx_restrict          ff,
 711                      t_forcerec                  * gmx_restrict          fr,
 712                      t_mdatoms                   * gmx_restrict     mdatoms,
 713                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 714                      t_nrnb                      * gmx_restrict        nrnb)
 715 {
 716     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 717      * just 0 for non-waters.
 718      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 719      * jnr indices corresponding to data put in the four positions in the SIMD register.
 720      */
 721     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 722     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 723     int              jnrA,jnrB,jnrC,jnrD;
 724     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 725     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 726     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 727     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 728     real             rcutoff_scalar;
 729     real             *shiftvec,*fshift,*x,*f;
 730     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 731     real             scratch[4*DIM];
 732     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 733     real *           vdwioffsetptr0;
 734     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 735     real *           vdwioffsetptr1;
 736     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 737     real *           vdwioffsetptr2;
 738     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 739     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 740     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 741     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 742     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 743     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 744     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 745     real             *charge;
 746     int              nvdwtype;
 747     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 748     int              *vdwtype;
 749     real             *vdwparam;
 750     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 751     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 752     __m128i          vfitab;
 753     __m128i          ifour       = _mm_set1_epi32(4);
 754     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 755     real             *vftab;
 756     __m256d          dummy_mask,cutoff_mask;
 757     __m128           tmpmask0,tmpmask1;
 758     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 759     __m256d          one     = _mm256_set1_pd(1.0);
 760     __m256d          two     = _mm256_set1_pd(2.0);
 761     x                = xx[0];
 762     f                = ff[0];
 763
 764     nri              = nlist->nri;
 765     iinr             = nlist->iinr;
 766     jindex           = nlist->jindex;
 767     jjnr             = nlist->jjnr;
 768     shiftidx         = nlist->shift;
 769     gid              = nlist->gid;
 770     shiftvec         = fr->shift_vec[0];
 771     fshift           = fr->fshift[0];
 772     facel            = _mm256_set1_pd(fr->epsfac);
 773     charge           = mdatoms->chargeA;
 774     krf              = _mm256_set1_pd(fr->ic->k_rf);
 775     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 776     crf              = _mm256_set1_pd(fr->ic->c_rf);
 777     nvdwtype         = fr->ntype;
 778     vdwparam         = fr->nbfp;
 779     vdwtype          = mdatoms->typeA;
 780
 781     vftab            = kernel_data->table_vdw->data;
 782     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 783
 784     /* Setup water-specific parameters */
 785     inr              = nlist->iinr[0];
 786     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 787     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 788     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 789     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 790
 791     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 792     rcutoff_scalar   = fr->rcoulomb;
 793     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 794     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 795
 796     /* Avoid stupid compiler warnings */
 797     jnrA = jnrB = jnrC = jnrD = 0;
 798     j_coord_offsetA = 0;
 799     j_coord_offsetB = 0;
 800     j_coord_offsetC = 0;
 801     j_coord_offsetD = 0;
 802
 803     outeriter        = 0;
 804     inneriter        = 0;
 805
 806     for(iidx=0;iidx<4*DIM;iidx++)
 807     {
 808         scratch[iidx] = 0.0;
 809     }
 810
 811     /* Start outer loop over neighborlists */
 812     for(iidx=0; iidx<nri; iidx++)
 813     {
 814         /* Load shift vector for this list */
 815         i_shift_offset   = DIM*shiftidx[iidx];
 816
 817         /* Load limits for loop over neighbors */
 818         j_index_start    = jindex[iidx];
 819         j_index_end      = jindex[iidx+1];
 820
 821         /* Get outer coordinate index */
 822         inr              = iinr[iidx];
 823         i_coord_offset   = DIM*inr;
 824
 825         /* Load i particle coords and add shift vector */
 826         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 827                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 828
 829         fix0             = _mm256_setzero_pd();
 830         fiy0             = _mm256_setzero_pd();
 831         fiz0             = _mm256_setzero_pd();
 832         fix1             = _mm256_setzero_pd();
 833         fiy1             = _mm256_setzero_pd();
 834         fiz1             = _mm256_setzero_pd();
 835         fix2             = _mm256_setzero_pd();
 836         fiy2             = _mm256_setzero_pd();
 837         fiz2             = _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
 869             /* Calculate squared distance and things based on it */
 870             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 871             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 872             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 873
 874             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 875             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 876             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 877
 878             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 879             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 880             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 884                                                                  charge+jnrC+0,charge+jnrD+0);
 885             vdwjidx0A        = 2*vdwtype[jnrA+0];
 886             vdwjidx0B        = 2*vdwtype[jnrB+0];
 887             vdwjidx0C        = 2*vdwtype[jnrC+0];
 888             vdwjidx0D        = 2*vdwtype[jnrD+0];
 889
 890             fjx0             = _mm256_setzero_pd();
 891             fjy0             = _mm256_setzero_pd();
 892             fjz0             = _mm256_setzero_pd();
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 899             {
 900
 901             r00              = _mm256_mul_pd(rsq00,rinv00);
 902
 903             /* Compute parameters for interactions between i and j atoms */
 904             qq00             = _mm256_mul_pd(iq0,jq0);
 905             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 906                                             vdwioffsetptr0+vdwjidx0B,
 907                                             vdwioffsetptr0+vdwjidx0C,
 908                                             vdwioffsetptr0+vdwjidx0D,
 909                                             &c6_00,&c12_00);
 910
 911             /* Calculate table index by multiplying r with table scale and truncate to integer */
 912             rt               = _mm256_mul_pd(r00,vftabscale);
 913             vfitab           = _mm256_cvttpd_epi32(rt);
 914             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 915             vfitab           = _mm_slli_epi32(vfitab,3);
 916
 917             /* REACTION-FIELD ELECTROSTATICS */
 918             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 919
 920             /* CUBIC SPLINE TABLE DISPERSION */
 921             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 922             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 923             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 924             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 925             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 926             Heps             = _mm256_mul_pd(vfeps,H);
 927             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 928             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 929             fvdw6            = _mm256_mul_pd(c6_00,FF);
 930
 931             /* CUBIC SPLINE TABLE REPULSION */
 932             vfitab           = _mm_add_epi32(vfitab,ifour);
 933             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 934             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 935             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 936             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 937             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 938             Heps             = _mm256_mul_pd(vfeps,H);
 939             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 940             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 941             fvdw12           = _mm256_mul_pd(c12_00,FF);
 942             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 943
 944             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 945
 946             fscal            = _mm256_add_pd(felec,fvdw);
 947
 948             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_pd(fscal,dx00);
 952             ty               = _mm256_mul_pd(fscal,dy00);
 953             tz               = _mm256_mul_pd(fscal,dz00);
 954
 955             /* Update vectorial force */
 956             fix0             = _mm256_add_pd(fix0,tx);
 957             fiy0             = _mm256_add_pd(fiy0,ty);
 958             fiz0             = _mm256_add_pd(fiz0,tz);
 959
 960             fjx0             = _mm256_add_pd(fjx0,tx);
 961             fjy0             = _mm256_add_pd(fjy0,ty);
 962             fjz0             = _mm256_add_pd(fjz0,tz);
 963
 964             }
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 971             {
 972
 973             /* Compute parameters for interactions between i and j atoms */
 974             qq10             = _mm256_mul_pd(iq1,jq0);
 975
 976             /* REACTION-FIELD ELECTROSTATICS */
 977             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 978
 979             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 980
 981             fscal            = felec;
 982
 983             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 984
 985             /* Calculate temporary vectorial force */
 986             tx               = _mm256_mul_pd(fscal,dx10);
 987             ty               = _mm256_mul_pd(fscal,dy10);
 988             tz               = _mm256_mul_pd(fscal,dz10);
 989
 990             /* Update vectorial force */
 991             fix1             = _mm256_add_pd(fix1,tx);
 992             fiy1             = _mm256_add_pd(fiy1,ty);
 993             fiz1             = _mm256_add_pd(fiz1,tz);
 994
 995             fjx0             = _mm256_add_pd(fjx0,tx);
 996             fjy0             = _mm256_add_pd(fjy0,ty);
 997             fjz0             = _mm256_add_pd(fjz0,tz);
 998
 999             }
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1006             {
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq20             = _mm256_mul_pd(iq2,jq0);
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1013
1014             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm256_mul_pd(fscal,dx20);
1022             ty               = _mm256_mul_pd(fscal,dy20);
1023             tz               = _mm256_mul_pd(fscal,dz20);
1024
1025             /* Update vectorial force */
1026             fix2             = _mm256_add_pd(fix2,tx);
1027             fiy2             = _mm256_add_pd(fiy2,ty);
1028             fiz2             = _mm256_add_pd(fiz2,tz);
1029
1030             fjx0             = _mm256_add_pd(fjx0,tx);
1031             fjy0             = _mm256_add_pd(fjy0,ty);
1032             fjz0             = _mm256_add_pd(fjz0,tz);
1033
1034             }
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 120 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
1089             /* Calculate squared distance and things based on it */
1090             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1091             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1092             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1093
1094             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1095             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1096             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1097
1098             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1099             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1100             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1101
1102             /* Load parameters for j particles */
1103             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1104                                                                  charge+jnrC+0,charge+jnrD+0);
1105             vdwjidx0A        = 2*vdwtype[jnrA+0];
1106             vdwjidx0B        = 2*vdwtype[jnrB+0];
1107             vdwjidx0C        = 2*vdwtype[jnrC+0];
1108             vdwjidx0D        = 2*vdwtype[jnrD+0];
1109
1110             fjx0             = _mm256_setzero_pd();
1111             fjy0             = _mm256_setzero_pd();
1112             fjz0             = _mm256_setzero_pd();
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1119             {
1120
1121             r00              = _mm256_mul_pd(rsq00,rinv00);
1122             r00              = _mm256_andnot_pd(dummy_mask,r00);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq00             = _mm256_mul_pd(iq0,jq0);
1126             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1127                                             vdwioffsetptr0+vdwjidx0B,
1128                                             vdwioffsetptr0+vdwjidx0C,
1129                                             vdwioffsetptr0+vdwjidx0D,
1130                                             &c6_00,&c12_00);
1131
1132             /* Calculate table index by multiplying r with table scale and truncate to integer */
1133             rt               = _mm256_mul_pd(r00,vftabscale);
1134             vfitab           = _mm256_cvttpd_epi32(rt);
1135             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1136             vfitab           = _mm_slli_epi32(vfitab,3);
1137
1138             /* REACTION-FIELD ELECTROSTATICS */
1139             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1140
1141             /* CUBIC SPLINE TABLE DISPERSION */
1142             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1143             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1144             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1145             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1146             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1147             Heps             = _mm256_mul_pd(vfeps,H);
1148             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1149             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1150             fvdw6            = _mm256_mul_pd(c6_00,FF);
1151
1152             /* CUBIC SPLINE TABLE REPULSION */
1153             vfitab           = _mm_add_epi32(vfitab,ifour);
1154             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1155             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1156             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1157             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1158             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1159             Heps             = _mm256_mul_pd(vfeps,H);
1160             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1161             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1162             fvdw12           = _mm256_mul_pd(c12_00,FF);
1163             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1164
1165             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1166
1167             fscal            = _mm256_add_pd(felec,fvdw);
1168
1169             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1170
1171             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm256_mul_pd(fscal,dx00);
1175             ty               = _mm256_mul_pd(fscal,dy00);
1176             tz               = _mm256_mul_pd(fscal,dz00);
1177
1178             /* Update vectorial force */
1179             fix0             = _mm256_add_pd(fix0,tx);
1180             fiy0             = _mm256_add_pd(fiy0,ty);
1181             fiz0             = _mm256_add_pd(fiz0,tz);
1182
1183             fjx0             = _mm256_add_pd(fjx0,tx);
1184             fjy0             = _mm256_add_pd(fjy0,ty);
1185             fjz0             = _mm256_add_pd(fjz0,tz);
1186
1187             }
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1194             {
1195
1196             /* Compute parameters for interactions between i and j atoms */
1197             qq10             = _mm256_mul_pd(iq1,jq0);
1198
1199             /* REACTION-FIELD ELECTROSTATICS */
1200             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1201
1202             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1207
1208             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1209
1210             /* Calculate temporary vectorial force */
1211             tx               = _mm256_mul_pd(fscal,dx10);
1212             ty               = _mm256_mul_pd(fscal,dy10);
1213             tz               = _mm256_mul_pd(fscal,dz10);
1214
1215             /* Update vectorial force */
1216             fix1             = _mm256_add_pd(fix1,tx);
1217             fiy1             = _mm256_add_pd(fiy1,ty);
1218             fiz1             = _mm256_add_pd(fiz1,tz);
1219
1220             fjx0             = _mm256_add_pd(fjx0,tx);
1221             fjy0             = _mm256_add_pd(fjy0,ty);
1222             fjz0             = _mm256_add_pd(fjz0,tz);
1223
1224             }
1225
1226             /**************************
1227              * CALCULATE INTERACTIONS *
1228              **************************/
1229
1230             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1231             {
1232
1233             /* Compute parameters for interactions between i and j atoms */
1234             qq20             = _mm256_mul_pd(iq2,jq0);
1235
1236             /* REACTION-FIELD ELECTROSTATICS */
1237             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1238
1239             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1240
1241             fscal            = felec;
1242
1243             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1244
1245             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm256_mul_pd(fscal,dx20);
1249             ty               = _mm256_mul_pd(fscal,dy20);
1250             tz               = _mm256_mul_pd(fscal,dz20);
1251
1252             /* Update vectorial force */
1253             fix2             = _mm256_add_pd(fix2,tx);
1254             fiy2             = _mm256_add_pd(fiy2,ty);
1255             fiz2             = _mm256_add_pd(fiz2,tz);
1256
1257             fjx0             = _mm256_add_pd(fjx0,tx);
1258             fjy0             = _mm256_add_pd(fjy0,ty);
1259             fjz0             = _mm256_add_pd(fjz0,tz);
1260
1261             }
1262
1263             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1264             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1265             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1266             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1267
1268             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1269
1270             /* Inner loop uses 121 flops */
1271         }
1272
1273         /* End of innermost loop */
1274
1275         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1276                                                  f+i_coord_offset,fshift+i_shift_offset);
1277
1278         /* Increment number of inner iterations */
1279         inneriter                  += j_index_end - j_index_start;
1280
1281         /* Outer loop uses 18 flops */
1282     }
1283
1284     /* Increment number of outer iterations */
1285     outeriter        += nri;
1286
1287     /* Update outer/inner flops */
1288
1289     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1290 }