src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     int              nvdwtype;
  84     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  85     int              *vdwtype;
  86     real             *vdwparam;
  87     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  88     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  89     __m128i          vfitab;
  90     __m128i          ifour       = _mm_set1_epi32(4);
  91     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  92     real             *vftab;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     krf              = _mm256_set1_pd(fr->ic->k_rf);
 112     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 113     crf              = _mm256_set1_pd(fr->ic->c_rf);
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_vdw->data;
 119     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 124     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 125     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 126     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 127
 128     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 129     rcutoff_scalar   = fr->rcoulomb;
 130     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 131     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 165
 166         fix0             = _mm256_setzero_pd();
 167         fiy0             = _mm256_setzero_pd();
 168         fiz0             = _mm256_setzero_pd();
 169         fix1             = _mm256_setzero_pd();
 170         fiy1             = _mm256_setzero_pd();
 171         fiz1             = _mm256_setzero_pd();
 172         fix2             = _mm256_setzero_pd();
 173         fiy2             = _mm256_setzero_pd();
 174         fiz2             = _mm256_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm256_setzero_pd();
 178         vvdwsum          = _mm256_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 197                                                  &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm256_sub_pd(ix0,jx0);
 201             dy00             = _mm256_sub_pd(iy0,jy0);
 202             dz00             = _mm256_sub_pd(iz0,jz0);
 203             dx10             = _mm256_sub_pd(ix1,jx0);
 204             dy10             = _mm256_sub_pd(iy1,jy0);
 205             dz10             = _mm256_sub_pd(iz1,jz0);
 206             dx20             = _mm256_sub_pd(ix2,jx0);
 207             dy20             = _mm256_sub_pd(iy2,jy0);
 208             dz20             = _mm256_sub_pd(iz2,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 214
 215             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 216             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 217             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 218
 219             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 220             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 221             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 225                                                                  charge+jnrC+0,charge+jnrD+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228             vdwjidx0C        = 2*vdwtype[jnrC+0];
 229             vdwjidx0D        = 2*vdwtype[jnrD+0];
 230
 231             fjx0             = _mm256_setzero_pd();
 232             fjy0             = _mm256_setzero_pd();
 233             fjz0             = _mm256_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 240             {
 241
 242             r00              = _mm256_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm256_mul_pd(iq0,jq0);
 246             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 247                                             vdwioffsetptr0+vdwjidx0B,
 248                                             vdwioffsetptr0+vdwjidx0C,
 249                                             vdwioffsetptr0+vdwjidx0D,
 250                                             &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm256_mul_pd(r00,vftabscale);
 254             vfitab           = _mm256_cvttpd_epi32(rt);
 255             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 256             vfitab           = _mm_slli_epi32(vfitab,3);
 257
 258             /* REACTION-FIELD ELECTROSTATICS */
 259             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 260             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 261
 262             /* CUBIC SPLINE TABLE DISPERSION */
 263             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 264             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 265             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 266             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 267             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 268             Heps             = _mm256_mul_pd(vfeps,H);
 269             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 270             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 271             vvdw6            = _mm256_mul_pd(c6_00,VV);
 272             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 273             fvdw6            = _mm256_mul_pd(c6_00,FF);
 274
 275             /* CUBIC SPLINE TABLE REPULSION */
 276             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 286             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 287             fvdw12           = _mm256_mul_pd(c12_00,FF);
 288             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 289             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 290
 291             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm256_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 297             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 298
 299             fscal            = _mm256_add_pd(felec,fvdw);
 300
 301             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm256_mul_pd(fscal,dx00);
 305             ty               = _mm256_mul_pd(fscal,dy00);
 306             tz               = _mm256_mul_pd(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm256_add_pd(fix0,tx);
 310             fiy0             = _mm256_add_pd(fiy0,ty);
 311             fiz0             = _mm256_add_pd(fiz0,tz);
 312
 313             fjx0             = _mm256_add_pd(fjx0,tx);
 314             fjy0             = _mm256_add_pd(fjy0,ty);
 315             fjz0             = _mm256_add_pd(fjz0,tz);
 316
 317             }
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 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             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm256_and_pd(velec,cutoff_mask);
 337             velecsum         = _mm256_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm256_mul_pd(fscal,dx10);
 345             ty               = _mm256_mul_pd(fscal,dy10);
 346             tz               = _mm256_mul_pd(fscal,dz10);
 347
 348             /* Update vectorial force */
 349             fix1             = _mm256_add_pd(fix1,tx);
 350             fiy1             = _mm256_add_pd(fiy1,ty);
 351             fiz1             = _mm256_add_pd(fiz1,tz);
 352
 353             fjx0             = _mm256_add_pd(fjx0,tx);
 354             fjy0             = _mm256_add_pd(fjy0,ty);
 355             fjz0             = _mm256_add_pd(fjz0,tz);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 364             {
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm256_mul_pd(iq2,jq0);
 368
 369             /* REACTION-FIELD ELECTROSTATICS */
 370             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 371             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 372
 373             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm256_and_pd(velec,cutoff_mask);
 377             velecsum         = _mm256_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm256_mul_pd(fscal,dx20);
 385             ty               = _mm256_mul_pd(fscal,dy20);
 386             tz               = _mm256_mul_pd(fscal,dz20);
 387
 388             /* Update vectorial force */
 389             fix2             = _mm256_add_pd(fix2,tx);
 390             fiy2             = _mm256_add_pd(fiy2,ty);
 391             fiz2             = _mm256_add_pd(fiz2,tz);
 392
 393             fjx0             = _mm256_add_pd(fjx0,tx);
 394             fjy0             = _mm256_add_pd(fjy0,ty);
 395             fjz0             = _mm256_add_pd(fjz0,tz);
 396
 397             }
 398
 399             fjptrA             = f+j_coord_offsetA;
 400             fjptrB             = f+j_coord_offsetB;
 401             fjptrC             = f+j_coord_offsetC;
 402             fjptrD             = f+j_coord_offsetD;
 403
 404             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 405
 406             /* Inner loop uses 147 flops */
 407         }
 408
 409         if(jidx<j_index_end)
 410         {
 411
 412             /* Get j neighbor index, and coordinate index */
 413             jnrlistA         = jjnr[jidx];
 414             jnrlistB         = jjnr[jidx+1];
 415             jnrlistC         = jjnr[jidx+2];
 416             jnrlistD         = jjnr[jidx+3];
 417             /* Sign of each element will be negative for non-real atoms.
 418              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 419              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 420              */
 421             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 422
 423             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 424             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 425             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 426
 427             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 428             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 429             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 430             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 431             j_coord_offsetA  = DIM*jnrA;
 432             j_coord_offsetB  = DIM*jnrB;
 433             j_coord_offsetC  = DIM*jnrC;
 434             j_coord_offsetD  = DIM*jnrD;
 435
 436             /* load j atom coordinates */
 437             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 438                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 439                                                  &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm256_sub_pd(ix0,jx0);
 443             dy00             = _mm256_sub_pd(iy0,jy0);
 444             dz00             = _mm256_sub_pd(iz0,jz0);
 445             dx10             = _mm256_sub_pd(ix1,jx0);
 446             dy10             = _mm256_sub_pd(iy1,jy0);
 447             dz10             = _mm256_sub_pd(iz1,jz0);
 448             dx20             = _mm256_sub_pd(ix2,jx0);
 449             dy20             = _mm256_sub_pd(iy2,jy0);
 450             dz20             = _mm256_sub_pd(iz2,jz0);
 451
 452             /* Calculate squared distance and things based on it */
 453             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 454             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 455             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 456
 457             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 458             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 459             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 460
 461             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 462             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 463             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 464
 465             /* Load parameters for j particles */
 466             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 467                                                                  charge+jnrC+0,charge+jnrD+0);
 468             vdwjidx0A        = 2*vdwtype[jnrA+0];
 469             vdwjidx0B        = 2*vdwtype[jnrB+0];
 470             vdwjidx0C        = 2*vdwtype[jnrC+0];
 471             vdwjidx0D        = 2*vdwtype[jnrD+0];
 472
 473             fjx0             = _mm256_setzero_pd();
 474             fjy0             = _mm256_setzero_pd();
 475             fjz0             = _mm256_setzero_pd();
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 482             {
 483
 484             r00              = _mm256_mul_pd(rsq00,rinv00);
 485             r00              = _mm256_andnot_pd(dummy_mask,r00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _mm256_mul_pd(iq0,jq0);
 489             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 490                                             vdwioffsetptr0+vdwjidx0B,
 491                                             vdwioffsetptr0+vdwjidx0C,
 492                                             vdwioffsetptr0+vdwjidx0D,
 493                                             &c6_00,&c12_00);
 494
 495             /* Calculate table index by multiplying r with table scale and truncate to integer */
 496             rt               = _mm256_mul_pd(r00,vftabscale);
 497             vfitab           = _mm256_cvttpd_epi32(rt);
 498             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 499             vfitab           = _mm_slli_epi32(vfitab,3);
 500
 501             /* REACTION-FIELD ELECTROSTATICS */
 502             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 503             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 504
 505             /* CUBIC SPLINE TABLE DISPERSION */
 506             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 507             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 508             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 509             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 510             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 511             Heps             = _mm256_mul_pd(vfeps,H);
 512             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 513             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 514             vvdw6            = _mm256_mul_pd(c6_00,VV);
 515             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 516             fvdw6            = _mm256_mul_pd(c6_00,FF);
 517
 518             /* CUBIC SPLINE TABLE REPULSION */
 519             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 529             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 530             fvdw12           = _mm256_mul_pd(c12_00,FF);
 531             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 532             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 533
 534             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm256_and_pd(velec,cutoff_mask);
 538             velec            = _mm256_andnot_pd(dummy_mask,velec);
 539             velecsum         = _mm256_add_pd(velecsum,velec);
 540             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 541             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 542             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 543
 544             fscal            = _mm256_add_pd(felec,fvdw);
 545
 546             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 547
 548             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 549
 550             /* Calculate temporary vectorial force */
 551             tx               = _mm256_mul_pd(fscal,dx00);
 552             ty               = _mm256_mul_pd(fscal,dy00);
 553             tz               = _mm256_mul_pd(fscal,dz00);
 554
 555             /* Update vectorial force */
 556             fix0             = _mm256_add_pd(fix0,tx);
 557             fiy0             = _mm256_add_pd(fiy0,ty);
 558             fiz0             = _mm256_add_pd(fiz0,tz);
 559
 560             fjx0             = _mm256_add_pd(fjx0,tx);
 561             fjy0             = _mm256_add_pd(fjy0,ty);
 562             fjz0             = _mm256_add_pd(fjz0,tz);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 571             {
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq10             = _mm256_mul_pd(iq1,jq0);
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 578             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 579
 580             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_and_pd(velec,cutoff_mask);
 584             velec            = _mm256_andnot_pd(dummy_mask,velec);
 585             velecsum         = _mm256_add_pd(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 590
 591             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_pd(fscal,dx10);
 595             ty               = _mm256_mul_pd(fscal,dy10);
 596             tz               = _mm256_mul_pd(fscal,dz10);
 597
 598             /* Update vectorial force */
 599             fix1             = _mm256_add_pd(fix1,tx);
 600             fiy1             = _mm256_add_pd(fiy1,ty);
 601             fiz1             = _mm256_add_pd(fiz1,tz);
 602
 603             fjx0             = _mm256_add_pd(fjx0,tx);
 604             fjy0             = _mm256_add_pd(fjy0,ty);
 605             fjz0             = _mm256_add_pd(fjz0,tz);
 606
 607             }
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 614             {
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq20             = _mm256_mul_pd(iq2,jq0);
 618
 619             /* REACTION-FIELD ELECTROSTATICS */
 620             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 621             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 622
 623             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velec            = _mm256_and_pd(velec,cutoff_mask);
 627             velec            = _mm256_andnot_pd(dummy_mask,velec);
 628             velecsum         = _mm256_add_pd(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 633
 634             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_pd(fscal,dx20);
 638             ty               = _mm256_mul_pd(fscal,dy20);
 639             tz               = _mm256_mul_pd(fscal,dz20);
 640
 641             /* Update vectorial force */
 642             fix2             = _mm256_add_pd(fix2,tx);
 643             fiy2             = _mm256_add_pd(fiy2,ty);
 644             fiz2             = _mm256_add_pd(fiz2,tz);
 645
 646             fjx0             = _mm256_add_pd(fjx0,tx);
 647             fjy0             = _mm256_add_pd(fjy0,ty);
 648             fjz0             = _mm256_add_pd(fjz0,tz);
 649
 650             }
 651
 652             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 653             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 654             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 655             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 656
 657             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 658
 659             /* Inner loop uses 148 flops */
 660         }
 661
 662         /* End of innermost loop */
 663
 664         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 665                                                  f+i_coord_offset,fshift+i_shift_offset);
 666
 667         ggid                        = gid[iidx];
 668         /* Update potential energies */
 669         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 670         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 671
 672         /* Increment number of inner iterations */
 673         inneriter                  += j_index_end - j_index_start;
 674
 675         /* Outer loop uses 20 flops */
 676     }
 677
 678     /* Increment number of outer iterations */
 679     outeriter        += nri;
 680
 681     /* Update outer/inner flops */
 682
 683     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 684 }
 685 /*
 686  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 687  * Electrostatics interaction: ReactionField
 688  * VdW interaction:            CubicSplineTable
 689  * Geometry:                   Water3-Particle
 690  * Calculate force/pot:        Force
 691  */
 692 void
 693 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 694                     (t_nblist * gmx_restrict                nlist,
 695                      rvec * gmx_restrict                    xx,
 696                      rvec * gmx_restrict                    ff,
 697                      t_forcerec * gmx_restrict              fr,
 698                      t_mdatoms * gmx_restrict               mdatoms,
 699                      nb_kernel_data_t * gmx_restrict        kernel_data,
 700                      t_nrnb * gmx_restrict                  nrnb)
 701 {
 702     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 703      * just 0 for non-waters.
 704      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 705      * jnr indices corresponding to data put in the four positions in the SIMD register.
 706      */
 707     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 708     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 709     int              jnrA,jnrB,jnrC,jnrD;
 710     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 711     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 712     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 713     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 714     real             rcutoff_scalar;
 715     real             *shiftvec,*fshift,*x,*f;
 716     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 717     real             scratch[4*DIM];
 718     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 719     real *           vdwioffsetptr0;
 720     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 721     real *           vdwioffsetptr1;
 722     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 723     real *           vdwioffsetptr2;
 724     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 725     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 726     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 727     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 728     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 729     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 730     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 731     real             *charge;
 732     int              nvdwtype;
 733     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 734     int              *vdwtype;
 735     real             *vdwparam;
 736     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 737     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 738     __m128i          vfitab;
 739     __m128i          ifour       = _mm_set1_epi32(4);
 740     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 741     real             *vftab;
 742     __m256d          dummy_mask,cutoff_mask;
 743     __m128           tmpmask0,tmpmask1;
 744     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 745     __m256d          one     = _mm256_set1_pd(1.0);
 746     __m256d          two     = _mm256_set1_pd(2.0);
 747     x                = xx[0];
 748     f                = ff[0];
 749
 750     nri              = nlist->nri;
 751     iinr             = nlist->iinr;
 752     jindex           = nlist->jindex;
 753     jjnr             = nlist->jjnr;
 754     shiftidx         = nlist->shift;
 755     gid              = nlist->gid;
 756     shiftvec         = fr->shift_vec[0];
 757     fshift           = fr->fshift[0];
 758     facel            = _mm256_set1_pd(fr->epsfac);
 759     charge           = mdatoms->chargeA;
 760     krf              = _mm256_set1_pd(fr->ic->k_rf);
 761     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 762     crf              = _mm256_set1_pd(fr->ic->c_rf);
 763     nvdwtype         = fr->ntype;
 764     vdwparam         = fr->nbfp;
 765     vdwtype          = mdatoms->typeA;
 766
 767     vftab            = kernel_data->table_vdw->data;
 768     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 769
 770     /* Setup water-specific parameters */
 771     inr              = nlist->iinr[0];
 772     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 773     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 774     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 775     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 776
 777     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 778     rcutoff_scalar   = fr->rcoulomb;
 779     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 780     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 781
 782     /* Avoid stupid compiler warnings */
 783     jnrA = jnrB = jnrC = jnrD = 0;
 784     j_coord_offsetA = 0;
 785     j_coord_offsetB = 0;
 786     j_coord_offsetC = 0;
 787     j_coord_offsetD = 0;
 788
 789     outeriter        = 0;
 790     inneriter        = 0;
 791
 792     for(iidx=0;iidx<4*DIM;iidx++)
 793     {
 794         scratch[iidx] = 0.0;
 795     }
 796
 797     /* Start outer loop over neighborlists */
 798     for(iidx=0; iidx<nri; iidx++)
 799     {
 800         /* Load shift vector for this list */
 801         i_shift_offset   = DIM*shiftidx[iidx];
 802
 803         /* Load limits for loop over neighbors */
 804         j_index_start    = jindex[iidx];
 805         j_index_end      = jindex[iidx+1];
 806
 807         /* Get outer coordinate index */
 808         inr              = iinr[iidx];
 809         i_coord_offset   = DIM*inr;
 810
 811         /* Load i particle coords and add shift vector */
 812         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 813                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 814
 815         fix0             = _mm256_setzero_pd();
 816         fiy0             = _mm256_setzero_pd();
 817         fiz0             = _mm256_setzero_pd();
 818         fix1             = _mm256_setzero_pd();
 819         fiy1             = _mm256_setzero_pd();
 820         fiz1             = _mm256_setzero_pd();
 821         fix2             = _mm256_setzero_pd();
 822         fiy2             = _mm256_setzero_pd();
 823         fiz2             = _mm256_setzero_pd();
 824
 825         /* Start inner kernel loop */
 826         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 827         {
 828
 829             /* Get j neighbor index, and coordinate index */
 830             jnrA             = jjnr[jidx];
 831             jnrB             = jjnr[jidx+1];
 832             jnrC             = jjnr[jidx+2];
 833             jnrD             = jjnr[jidx+3];
 834             j_coord_offsetA  = DIM*jnrA;
 835             j_coord_offsetB  = DIM*jnrB;
 836             j_coord_offsetC  = DIM*jnrC;
 837             j_coord_offsetD  = DIM*jnrD;
 838
 839             /* load j atom coordinates */
 840             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 841                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 842                                                  &jx0,&jy0,&jz0);
 843
 844             /* Calculate displacement vector */
 845             dx00             = _mm256_sub_pd(ix0,jx0);
 846             dy00             = _mm256_sub_pd(iy0,jy0);
 847             dz00             = _mm256_sub_pd(iz0,jz0);
 848             dx10             = _mm256_sub_pd(ix1,jx0);
 849             dy10             = _mm256_sub_pd(iy1,jy0);
 850             dz10             = _mm256_sub_pd(iz1,jz0);
 851             dx20             = _mm256_sub_pd(ix2,jx0);
 852             dy20             = _mm256_sub_pd(iy2,jy0);
 853             dz20             = _mm256_sub_pd(iz2,jz0);
 854
 855             /* Calculate squared distance and things based on it */
 856             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 857             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 858             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 859
 860             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 861             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 862             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 863
 864             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 865             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 866             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 867
 868             /* Load parameters for j particles */
 869             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 870                                                                  charge+jnrC+0,charge+jnrD+0);
 871             vdwjidx0A        = 2*vdwtype[jnrA+0];
 872             vdwjidx0B        = 2*vdwtype[jnrB+0];
 873             vdwjidx0C        = 2*vdwtype[jnrC+0];
 874             vdwjidx0D        = 2*vdwtype[jnrD+0];
 875
 876             fjx0             = _mm256_setzero_pd();
 877             fjy0             = _mm256_setzero_pd();
 878             fjz0             = _mm256_setzero_pd();
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 885             {
 886
 887             r00              = _mm256_mul_pd(rsq00,rinv00);
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq00             = _mm256_mul_pd(iq0,jq0);
 891             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 892                                             vdwioffsetptr0+vdwjidx0B,
 893                                             vdwioffsetptr0+vdwjidx0C,
 894                                             vdwioffsetptr0+vdwjidx0D,
 895                                             &c6_00,&c12_00);
 896
 897             /* Calculate table index by multiplying r with table scale and truncate to integer */
 898             rt               = _mm256_mul_pd(r00,vftabscale);
 899             vfitab           = _mm256_cvttpd_epi32(rt);
 900             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 901             vfitab           = _mm_slli_epi32(vfitab,3);
 902
 903             /* REACTION-FIELD ELECTROSTATICS */
 904             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 905
 906             /* CUBIC SPLINE TABLE DISPERSION */
 907             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 908             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 909             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 910             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 911             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 912             Heps             = _mm256_mul_pd(vfeps,H);
 913             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 914             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 915             fvdw6            = _mm256_mul_pd(c6_00,FF);
 916
 917             /* CUBIC SPLINE TABLE REPULSION */
 918             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm256_mul_pd(c12_00,FF);
 928             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 929
 930             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 931
 932             fscal            = _mm256_add_pd(felec,fvdw);
 933
 934             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm256_mul_pd(fscal,dx00);
 938             ty               = _mm256_mul_pd(fscal,dy00);
 939             tz               = _mm256_mul_pd(fscal,dz00);
 940
 941             /* Update vectorial force */
 942             fix0             = _mm256_add_pd(fix0,tx);
 943             fiy0             = _mm256_add_pd(fiy0,ty);
 944             fiz0             = _mm256_add_pd(fiz0,tz);
 945
 946             fjx0             = _mm256_add_pd(fjx0,tx);
 947             fjy0             = _mm256_add_pd(fjy0,ty);
 948             fjz0             = _mm256_add_pd(fjz0,tz);
 949
 950             }
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 957             {
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq10             = _mm256_mul_pd(iq1,jq0);
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 964
 965             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 966
 967             fscal            = felec;
 968
 969             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm256_mul_pd(fscal,dx10);
 973             ty               = _mm256_mul_pd(fscal,dy10);
 974             tz               = _mm256_mul_pd(fscal,dz10);
 975
 976             /* Update vectorial force */
 977             fix1             = _mm256_add_pd(fix1,tx);
 978             fiy1             = _mm256_add_pd(fiy1,ty);
 979             fiz1             = _mm256_add_pd(fiz1,tz);
 980
 981             fjx0             = _mm256_add_pd(fjx0,tx);
 982             fjy0             = _mm256_add_pd(fjy0,ty);
 983             fjz0             = _mm256_add_pd(fjz0,tz);
 984
 985             }
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 992             {
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq20             = _mm256_mul_pd(iq2,jq0);
 996
 997             /* REACTION-FIELD ELECTROSTATICS */
 998             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 999
1000             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm256_mul_pd(fscal,dx20);
1008             ty               = _mm256_mul_pd(fscal,dy20);
1009             tz               = _mm256_mul_pd(fscal,dz20);
1010
1011             /* Update vectorial force */
1012             fix2             = _mm256_add_pd(fix2,tx);
1013             fiy2             = _mm256_add_pd(fiy2,ty);
1014             fiz2             = _mm256_add_pd(fiz2,tz);
1015
1016             fjx0             = _mm256_add_pd(fjx0,tx);
1017             fjy0             = _mm256_add_pd(fjy0,ty);
1018             fjz0             = _mm256_add_pd(fjz0,tz);
1019
1020             }
1021
1022             fjptrA             = f+j_coord_offsetA;
1023             fjptrB             = f+j_coord_offsetB;
1024             fjptrC             = f+j_coord_offsetC;
1025             fjptrD             = f+j_coord_offsetD;
1026
1027             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1028
1029             /* Inner loop uses 120 flops */
1030         }
1031
1032         if(jidx<j_index_end)
1033         {
1034
1035             /* Get j neighbor index, and coordinate index */
1036             jnrlistA         = jjnr[jidx];
1037             jnrlistB         = jjnr[jidx+1];
1038             jnrlistC         = jjnr[jidx+2];
1039             jnrlistD         = jjnr[jidx+3];
1040             /* Sign of each element will be negative for non-real atoms.
1041              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1042              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1043              */
1044             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1045
1046             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1047             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1048             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1049
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             j_coord_offsetA  = DIM*jnrA;
1055             j_coord_offsetB  = DIM*jnrB;
1056             j_coord_offsetC  = DIM*jnrC;
1057             j_coord_offsetD  = DIM*jnrD;
1058
1059             /* load j atom coordinates */
1060             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1061                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1062                                                  &jx0,&jy0,&jz0);
1063
1064             /* Calculate displacement vector */
1065             dx00             = _mm256_sub_pd(ix0,jx0);
1066             dy00             = _mm256_sub_pd(iy0,jy0);
1067             dz00             = _mm256_sub_pd(iz0,jz0);
1068             dx10             = _mm256_sub_pd(ix1,jx0);
1069             dy10             = _mm256_sub_pd(iy1,jy0);
1070             dz10             = _mm256_sub_pd(iz1,jz0);
1071             dx20             = _mm256_sub_pd(ix2,jx0);
1072             dy20             = _mm256_sub_pd(iy2,jy0);
1073             dz20             = _mm256_sub_pd(iz2,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1077             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1078             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1079
1080             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1081             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1082             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1083
1084             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1085             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1086             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1087
1088             /* Load parameters for j particles */
1089             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1090                                                                  charge+jnrC+0,charge+jnrD+0);
1091             vdwjidx0A        = 2*vdwtype[jnrA+0];
1092             vdwjidx0B        = 2*vdwtype[jnrB+0];
1093             vdwjidx0C        = 2*vdwtype[jnrC+0];
1094             vdwjidx0D        = 2*vdwtype[jnrD+0];
1095
1096             fjx0             = _mm256_setzero_pd();
1097             fjy0             = _mm256_setzero_pd();
1098             fjz0             = _mm256_setzero_pd();
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1105             {
1106
1107             r00              = _mm256_mul_pd(rsq00,rinv00);
1108             r00              = _mm256_andnot_pd(dummy_mask,r00);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq00             = _mm256_mul_pd(iq0,jq0);
1112             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1113                                             vdwioffsetptr0+vdwjidx0B,
1114                                             vdwioffsetptr0+vdwjidx0C,
1115                                             vdwioffsetptr0+vdwjidx0D,
1116                                             &c6_00,&c12_00);
1117
1118             /* Calculate table index by multiplying r with table scale and truncate to integer */
1119             rt               = _mm256_mul_pd(r00,vftabscale);
1120             vfitab           = _mm256_cvttpd_epi32(rt);
1121             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1122             vfitab           = _mm_slli_epi32(vfitab,3);
1123
1124             /* REACTION-FIELD ELECTROSTATICS */
1125             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1126
1127             /* CUBIC SPLINE TABLE DISPERSION */
1128             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1129             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1130             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1131             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1132             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1133             Heps             = _mm256_mul_pd(vfeps,H);
1134             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1135             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1136             fvdw6            = _mm256_mul_pd(c6_00,FF);
1137
1138             /* CUBIC SPLINE TABLE REPULSION */
1139             vfitab           = _mm_add_epi32(vfitab,ifour);
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             fvdw12           = _mm256_mul_pd(c12_00,FF);
1149             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1150
1151             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1152
1153             fscal            = _mm256_add_pd(felec,fvdw);
1154
1155             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1156
1157             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1158
1159             /* Calculate temporary vectorial force */
1160             tx               = _mm256_mul_pd(fscal,dx00);
1161             ty               = _mm256_mul_pd(fscal,dy00);
1162             tz               = _mm256_mul_pd(fscal,dz00);
1163
1164             /* Update vectorial force */
1165             fix0             = _mm256_add_pd(fix0,tx);
1166             fiy0             = _mm256_add_pd(fiy0,ty);
1167             fiz0             = _mm256_add_pd(fiz0,tz);
1168
1169             fjx0             = _mm256_add_pd(fjx0,tx);
1170             fjy0             = _mm256_add_pd(fjy0,ty);
1171             fjz0             = _mm256_add_pd(fjz0,tz);
1172
1173             }
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1180             {
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm256_mul_pd(iq1,jq0);
1184
1185             /* REACTION-FIELD ELECTROSTATICS */
1186             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1187
1188             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1193
1194             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1195
1196             /* Calculate temporary vectorial force */
1197             tx               = _mm256_mul_pd(fscal,dx10);
1198             ty               = _mm256_mul_pd(fscal,dy10);
1199             tz               = _mm256_mul_pd(fscal,dz10);
1200
1201             /* Update vectorial force */
1202             fix1             = _mm256_add_pd(fix1,tx);
1203             fiy1             = _mm256_add_pd(fiy1,ty);
1204             fiz1             = _mm256_add_pd(fiz1,tz);
1205
1206             fjx0             = _mm256_add_pd(fjx0,tx);
1207             fjy0             = _mm256_add_pd(fjy0,ty);
1208             fjz0             = _mm256_add_pd(fjz0,tz);
1209
1210             }
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1217             {
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             qq20             = _mm256_mul_pd(iq2,jq0);
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1224
1225             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1230
1231             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm256_mul_pd(fscal,dx20);
1235             ty               = _mm256_mul_pd(fscal,dy20);
1236             tz               = _mm256_mul_pd(fscal,dz20);
1237
1238             /* Update vectorial force */
1239             fix2             = _mm256_add_pd(fix2,tx);
1240             fiy2             = _mm256_add_pd(fiy2,ty);
1241             fiz2             = _mm256_add_pd(fiz2,tz);
1242
1243             fjx0             = _mm256_add_pd(fjx0,tx);
1244             fjy0             = _mm256_add_pd(fjy0,ty);
1245             fjz0             = _mm256_add_pd(fjz0,tz);
1246
1247             }
1248
1249             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1250             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1251             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1252             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1253
1254             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1255
1256             /* Inner loop uses 121 flops */
1257         }
1258
1259         /* End of innermost loop */
1260
1261         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1262                                                  f+i_coord_offset,fshift+i_shift_offset);
1263
1264         /* Increment number of inner iterations */
1265         inneriter                  += j_index_end - j_index_start;
1266
1267         /* Outer loop uses 18 flops */
1268     }
1269
1270     /* Increment number of outer iterations */
1271     outeriter        += nri;
1272
1273     /* Update outer/inner flops */
1274
1275     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1276 }