src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSw_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_VdwLJSw_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_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     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  90     real             rswitch_scalar,d_scalar;
  91     __m256d          dummy_mask,cutoff_mask;
  92     __m128           tmpmask0,tmpmask1;
  93     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  94     __m256d          one     = _mm256_set1_pd(1.0);
  95     __m256d          two     = _mm256_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm256_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm256_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm256_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 119     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 120     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 121     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 126     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 127
 128     rswitch_scalar   = fr->rvdw_switch;
 129     rswitch          = _mm256_set1_pd(rswitch_scalar);
 130     /* Setup switch parameters */
 131     d_scalar         = rcutoff_scalar-rswitch_scalar;
 132     d                = _mm256_set1_pd(d_scalar);
 133     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 134     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 135     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 137     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm256_setzero_pd();
 174         fiy0             = _mm256_setzero_pd();
 175         fiz0             = _mm256_setzero_pd();
 176         fix1             = _mm256_setzero_pd();
 177         fiy1             = _mm256_setzero_pd();
 178         fiz1             = _mm256_setzero_pd();
 179         fix2             = _mm256_setzero_pd();
 180         fiy2             = _mm256_setzero_pd();
 181         fiz2             = _mm256_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_pd();
 185         vvdwsum          = _mm256_setzero_pd();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 204                                                  &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm256_sub_pd(ix0,jx0);
 208             dy00             = _mm256_sub_pd(iy0,jy0);
 209             dz00             = _mm256_sub_pd(iz0,jz0);
 210             dx10             = _mm256_sub_pd(ix1,jx0);
 211             dy10             = _mm256_sub_pd(iy1,jy0);
 212             dz10             = _mm256_sub_pd(iz1,jz0);
 213             dx20             = _mm256_sub_pd(ix2,jx0);
 214             dy20             = _mm256_sub_pd(iy2,jy0);
 215             dz20             = _mm256_sub_pd(iz2,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 219             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 220             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 221
 222             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 223             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 225
 226             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 227             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 228             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 232                                                                  charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             fjx0             = _mm256_setzero_pd();
 239             fjy0             = _mm256_setzero_pd();
 240             fjz0             = _mm256_setzero_pd();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 247             {
 248
 249             r00              = _mm256_mul_pd(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_pd(iq0,jq0);
 253             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             &c6_00,&c12_00);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 261             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 267             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 269             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 270
 271             d                = _mm256_sub_pd(r00,rswitch);
 272             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 273             d2               = _mm256_mul_pd(d,d);
 274             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 275
 276             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 277
 278             /* Evaluate switch function */
 279             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 280             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 281             vvdw             = _mm256_mul_pd(vvdw,sw);
 282             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velec            = _mm256_and_pd(velec,cutoff_mask);
 286             velecsum         = _mm256_add_pd(velecsum,velec);
 287             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 288             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = _mm256_add_pd(felec,fvdw);
 291
 292             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_pd(fscal,dx00);
 296             ty               = _mm256_mul_pd(fscal,dy00);
 297             tz               = _mm256_mul_pd(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm256_add_pd(fix0,tx);
 301             fiy0             = _mm256_add_pd(fiy0,ty);
 302             fiz0             = _mm256_add_pd(fiz0,tz);
 303
 304             fjx0             = _mm256_add_pd(fjx0,tx);
 305             fjy0             = _mm256_add_pd(fjy0,ty);
 306             fjz0             = _mm256_add_pd(fjz0,tz);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 315             {
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm256_mul_pd(iq1,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 322             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 323
 324             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm256_and_pd(velec,cutoff_mask);
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm256_mul_pd(fscal,dx10);
 336             ty               = _mm256_mul_pd(fscal,dy10);
 337             tz               = _mm256_mul_pd(fscal,dz10);
 338
 339             /* Update vectorial force */
 340             fix1             = _mm256_add_pd(fix1,tx);
 341             fiy1             = _mm256_add_pd(fiy1,ty);
 342             fiz1             = _mm256_add_pd(fiz1,tz);
 343
 344             fjx0             = _mm256_add_pd(fjx0,tx);
 345             fjy0             = _mm256_add_pd(fjy0,ty);
 346             fjz0             = _mm256_add_pd(fjz0,tz);
 347
 348             }
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 355             {
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm256_mul_pd(iq2,jq0);
 359
 360             /* REACTION-FIELD ELECTROSTATICS */
 361             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 362             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 363
 364             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm256_and_pd(velec,cutoff_mask);
 368             velecsum         = _mm256_add_pd(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_pd(fscal,dx20);
 376             ty               = _mm256_mul_pd(fscal,dy20);
 377             tz               = _mm256_mul_pd(fscal,dz20);
 378
 379             /* Update vectorial force */
 380             fix2             = _mm256_add_pd(fix2,tx);
 381             fiy2             = _mm256_add_pd(fiy2,ty);
 382             fiz2             = _mm256_add_pd(fiz2,tz);
 383
 384             fjx0             = _mm256_add_pd(fjx0,tx);
 385             fjy0             = _mm256_add_pd(fjy0,ty);
 386             fjz0             = _mm256_add_pd(fjz0,tz);
 387
 388             }
 389
 390             fjptrA             = f+j_coord_offsetA;
 391             fjptrB             = f+j_coord_offsetB;
 392             fjptrC             = f+j_coord_offsetC;
 393             fjptrD             = f+j_coord_offsetD;
 394
 395             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 396
 397             /* Inner loop uses 145 flops */
 398         }
 399
 400         if(jidx<j_index_end)
 401         {
 402
 403             /* Get j neighbor index, and coordinate index */
 404             jnrlistA         = jjnr[jidx];
 405             jnrlistB         = jjnr[jidx+1];
 406             jnrlistC         = jjnr[jidx+2];
 407             jnrlistD         = jjnr[jidx+3];
 408             /* Sign of each element will be negative for non-real atoms.
 409              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 410              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 411              */
 412             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 413
 414             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 415             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 416             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 417
 418             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 419             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 420             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 421             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 422             j_coord_offsetA  = DIM*jnrA;
 423             j_coord_offsetB  = DIM*jnrB;
 424             j_coord_offsetC  = DIM*jnrC;
 425             j_coord_offsetD  = DIM*jnrD;
 426
 427             /* load j atom coordinates */
 428             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 429                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 430                                                  &jx0,&jy0,&jz0);
 431
 432             /* Calculate displacement vector */
 433             dx00             = _mm256_sub_pd(ix0,jx0);
 434             dy00             = _mm256_sub_pd(iy0,jy0);
 435             dz00             = _mm256_sub_pd(iz0,jz0);
 436             dx10             = _mm256_sub_pd(ix1,jx0);
 437             dy10             = _mm256_sub_pd(iy1,jy0);
 438             dz10             = _mm256_sub_pd(iz1,jz0);
 439             dx20             = _mm256_sub_pd(ix2,jx0);
 440             dy20             = _mm256_sub_pd(iy2,jy0);
 441             dz20             = _mm256_sub_pd(iz2,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 445             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 446             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 447
 448             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 449             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 450             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 451
 452             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 453             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 454             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 458                                                                  charge+jnrC+0,charge+jnrD+0);
 459             vdwjidx0A        = 2*vdwtype[jnrA+0];
 460             vdwjidx0B        = 2*vdwtype[jnrB+0];
 461             vdwjidx0C        = 2*vdwtype[jnrC+0];
 462             vdwjidx0D        = 2*vdwtype[jnrD+0];
 463
 464             fjx0             = _mm256_setzero_pd();
 465             fjy0             = _mm256_setzero_pd();
 466             fjz0             = _mm256_setzero_pd();
 467
 468             /**************************
 469              * CALCULATE INTERACTIONS *
 470              **************************/
 471
 472             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 473             {
 474
 475             r00              = _mm256_mul_pd(rsq00,rinv00);
 476             r00              = _mm256_andnot_pd(dummy_mask,r00);
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             qq00             = _mm256_mul_pd(iq0,jq0);
 480             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 481                                             vdwioffsetptr0+vdwjidx0B,
 482                                             vdwioffsetptr0+vdwjidx0C,
 483                                             vdwioffsetptr0+vdwjidx0D,
 484                                             &c6_00,&c12_00);
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 488             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 489
 490             /* LENNARD-JONES DISPERSION/REPULSION */
 491
 492             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 493             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 494             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 495             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 496             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 497
 498             d                = _mm256_sub_pd(r00,rswitch);
 499             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 500             d2               = _mm256_mul_pd(d,d);
 501             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 502
 503             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 504
 505             /* Evaluate switch function */
 506             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 507             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 508             vvdw             = _mm256_mul_pd(vvdw,sw);
 509             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_and_pd(velec,cutoff_mask);
 513             velec            = _mm256_andnot_pd(dummy_mask,velec);
 514             velecsum         = _mm256_add_pd(velecsum,velec);
 515             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 516             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 517             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 518
 519             fscal            = _mm256_add_pd(felec,fvdw);
 520
 521             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 522
 523             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm256_mul_pd(fscal,dx00);
 527             ty               = _mm256_mul_pd(fscal,dy00);
 528             tz               = _mm256_mul_pd(fscal,dz00);
 529
 530             /* Update vectorial force */
 531             fix0             = _mm256_add_pd(fix0,tx);
 532             fiy0             = _mm256_add_pd(fiy0,ty);
 533             fiz0             = _mm256_add_pd(fiz0,tz);
 534
 535             fjx0             = _mm256_add_pd(fjx0,tx);
 536             fjy0             = _mm256_add_pd(fjy0,ty);
 537             fjz0             = _mm256_add_pd(fjz0,tz);
 538
 539             }
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 546             {
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq10             = _mm256_mul_pd(iq1,jq0);
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 553             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 554
 555             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm256_and_pd(velec,cutoff_mask);
 559             velec            = _mm256_andnot_pd(dummy_mask,velec);
 560             velecsum         = _mm256_add_pd(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 565
 566             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_pd(fscal,dx10);
 570             ty               = _mm256_mul_pd(fscal,dy10);
 571             tz               = _mm256_mul_pd(fscal,dz10);
 572
 573             /* Update vectorial force */
 574             fix1             = _mm256_add_pd(fix1,tx);
 575             fiy1             = _mm256_add_pd(fiy1,ty);
 576             fiz1             = _mm256_add_pd(fiz1,tz);
 577
 578             fjx0             = _mm256_add_pd(fjx0,tx);
 579             fjy0             = _mm256_add_pd(fjy0,ty);
 580             fjz0             = _mm256_add_pd(fjz0,tz);
 581
 582             }
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 589             {
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq20             = _mm256_mul_pd(iq2,jq0);
 593
 594             /* REACTION-FIELD ELECTROSTATICS */
 595             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 596             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 597
 598             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm256_and_pd(velec,cutoff_mask);
 602             velec            = _mm256_andnot_pd(dummy_mask,velec);
 603             velecsum         = _mm256_add_pd(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_pd(fscal,dx20);
 613             ty               = _mm256_mul_pd(fscal,dy20);
 614             tz               = _mm256_mul_pd(fscal,dz20);
 615
 616             /* Update vectorial force */
 617             fix2             = _mm256_add_pd(fix2,tx);
 618             fiy2             = _mm256_add_pd(fiy2,ty);
 619             fiz2             = _mm256_add_pd(fiz2,tz);
 620
 621             fjx0             = _mm256_add_pd(fjx0,tx);
 622             fjy0             = _mm256_add_pd(fjy0,ty);
 623             fjz0             = _mm256_add_pd(fjz0,tz);
 624
 625             }
 626
 627             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 628             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 629             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 630             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 631
 632             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 633
 634             /* Inner loop uses 146 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 640                                                  f+i_coord_offset,fshift+i_shift_offset);
 641
 642         ggid                        = gid[iidx];
 643         /* Update potential energies */
 644         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 645         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 646
 647         /* Increment number of inner iterations */
 648         inneriter                  += j_index_end - j_index_start;
 649
 650         /* Outer loop uses 20 flops */
 651     }
 652
 653     /* Increment number of outer iterations */
 654     outeriter        += nri;
 655
 656     /* Update outer/inner flops */
 657
 658     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 659 }
 660 /*
 661  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 662  * Electrostatics interaction: ReactionField
 663  * VdW interaction:            LennardJones
 664  * Geometry:                   Water3-Particle
 665  * Calculate force/pot:        Force
 666  */
 667 void
 668 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 669                     (t_nblist * gmx_restrict                nlist,
 670                      rvec * gmx_restrict                    xx,
 671                      rvec * gmx_restrict                    ff,
 672                      t_forcerec * gmx_restrict              fr,
 673                      t_mdatoms * gmx_restrict               mdatoms,
 674                      nb_kernel_data_t * gmx_restrict        kernel_data,
 675                      t_nrnb * gmx_restrict                  nrnb)
 676 {
 677     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 678      * just 0 for non-waters.
 679      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 680      * jnr indices corresponding to data put in the four positions in the SIMD register.
 681      */
 682     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 683     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 684     int              jnrA,jnrB,jnrC,jnrD;
 685     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 686     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 687     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 688     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 689     real             rcutoff_scalar;
 690     real             *shiftvec,*fshift,*x,*f;
 691     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 692     real             scratch[4*DIM];
 693     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 694     real *           vdwioffsetptr0;
 695     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 696     real *           vdwioffsetptr1;
 697     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 698     real *           vdwioffsetptr2;
 699     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 700     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 701     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 702     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 703     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 704     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 705     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 706     real             *charge;
 707     int              nvdwtype;
 708     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 709     int              *vdwtype;
 710     real             *vdwparam;
 711     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 712     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 713     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 714     real             rswitch_scalar,d_scalar;
 715     __m256d          dummy_mask,cutoff_mask;
 716     __m128           tmpmask0,tmpmask1;
 717     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 718     __m256d          one     = _mm256_set1_pd(1.0);
 719     __m256d          two     = _mm256_set1_pd(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm256_set1_pd(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     krf              = _mm256_set1_pd(fr->ic->k_rf);
 734     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 735     crf              = _mm256_set1_pd(fr->ic->c_rf);
 736     nvdwtype         = fr->ntype;
 737     vdwparam         = fr->nbfp;
 738     vdwtype          = mdatoms->typeA;
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 743     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 744     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 745     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 746
 747     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 748     rcutoff_scalar   = fr->rcoulomb;
 749     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 750     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 751
 752     rswitch_scalar   = fr->rvdw_switch;
 753     rswitch          = _mm256_set1_pd(rswitch_scalar);
 754     /* Setup switch parameters */
 755     d_scalar         = rcutoff_scalar-rswitch_scalar;
 756     d                = _mm256_set1_pd(d_scalar);
 757     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 758     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 759     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 760     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 761     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 762     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = jnrC = jnrD = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768     j_coord_offsetC = 0;
 769     j_coord_offsetD = 0;
 770
 771     outeriter        = 0;
 772     inneriter        = 0;
 773
 774     for(iidx=0;iidx<4*DIM;iidx++)
 775     {
 776         scratch[iidx] = 0.0;
 777     }
 778
 779     /* Start outer loop over neighborlists */
 780     for(iidx=0; iidx<nri; iidx++)
 781     {
 782         /* Load shift vector for this list */
 783         i_shift_offset   = DIM*shiftidx[iidx];
 784
 785         /* Load limits for loop over neighbors */
 786         j_index_start    = jindex[iidx];
 787         j_index_end      = jindex[iidx+1];
 788
 789         /* Get outer coordinate index */
 790         inr              = iinr[iidx];
 791         i_coord_offset   = DIM*inr;
 792
 793         /* Load i particle coords and add shift vector */
 794         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 795                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 796
 797         fix0             = _mm256_setzero_pd();
 798         fiy0             = _mm256_setzero_pd();
 799         fiz0             = _mm256_setzero_pd();
 800         fix1             = _mm256_setzero_pd();
 801         fiy1             = _mm256_setzero_pd();
 802         fiz1             = _mm256_setzero_pd();
 803         fix2             = _mm256_setzero_pd();
 804         fiy2             = _mm256_setzero_pd();
 805         fiz2             = _mm256_setzero_pd();
 806
 807         /* Start inner kernel loop */
 808         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 809         {
 810
 811             /* Get j neighbor index, and coordinate index */
 812             jnrA             = jjnr[jidx];
 813             jnrB             = jjnr[jidx+1];
 814             jnrC             = jjnr[jidx+2];
 815             jnrD             = jjnr[jidx+3];
 816             j_coord_offsetA  = DIM*jnrA;
 817             j_coord_offsetB  = DIM*jnrB;
 818             j_coord_offsetC  = DIM*jnrC;
 819             j_coord_offsetD  = DIM*jnrD;
 820
 821             /* load j atom coordinates */
 822             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 823                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 824                                                  &jx0,&jy0,&jz0);
 825
 826             /* Calculate displacement vector */
 827             dx00             = _mm256_sub_pd(ix0,jx0);
 828             dy00             = _mm256_sub_pd(iy0,jy0);
 829             dz00             = _mm256_sub_pd(iz0,jz0);
 830             dx10             = _mm256_sub_pd(ix1,jx0);
 831             dy10             = _mm256_sub_pd(iy1,jy0);
 832             dz10             = _mm256_sub_pd(iz1,jz0);
 833             dx20             = _mm256_sub_pd(ix2,jx0);
 834             dy20             = _mm256_sub_pd(iy2,jy0);
 835             dz20             = _mm256_sub_pd(iz2,jz0);
 836
 837             /* Calculate squared distance and things based on it */
 838             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 839             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 840             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 841
 842             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 843             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 844             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 845
 846             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 847             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 848             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 849
 850             /* Load parameters for j particles */
 851             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 852                                                                  charge+jnrC+0,charge+jnrD+0);
 853             vdwjidx0A        = 2*vdwtype[jnrA+0];
 854             vdwjidx0B        = 2*vdwtype[jnrB+0];
 855             vdwjidx0C        = 2*vdwtype[jnrC+0];
 856             vdwjidx0D        = 2*vdwtype[jnrD+0];
 857
 858             fjx0             = _mm256_setzero_pd();
 859             fjy0             = _mm256_setzero_pd();
 860             fjz0             = _mm256_setzero_pd();
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 867             {
 868
 869             r00              = _mm256_mul_pd(rsq00,rinv00);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq00             = _mm256_mul_pd(iq0,jq0);
 873             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 874                                             vdwioffsetptr0+vdwjidx0B,
 875                                             vdwioffsetptr0+vdwjidx0C,
 876                                             vdwioffsetptr0+vdwjidx0D,
 877                                             &c6_00,&c12_00);
 878
 879             /* REACTION-FIELD ELECTROSTATICS */
 880             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 881
 882             /* LENNARD-JONES DISPERSION/REPULSION */
 883
 884             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 885             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 886             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 887             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 888             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 889
 890             d                = _mm256_sub_pd(r00,rswitch);
 891             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 892             d2               = _mm256_mul_pd(d,d);
 893             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 894
 895             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 896
 897             /* Evaluate switch function */
 898             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 899             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 900             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 901
 902             fscal            = _mm256_add_pd(felec,fvdw);
 903
 904             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 905
 906             /* Calculate temporary vectorial force */
 907             tx               = _mm256_mul_pd(fscal,dx00);
 908             ty               = _mm256_mul_pd(fscal,dy00);
 909             tz               = _mm256_mul_pd(fscal,dz00);
 910
 911             /* Update vectorial force */
 912             fix0             = _mm256_add_pd(fix0,tx);
 913             fiy0             = _mm256_add_pd(fiy0,ty);
 914             fiz0             = _mm256_add_pd(fiz0,tz);
 915
 916             fjx0             = _mm256_add_pd(fjx0,tx);
 917             fjy0             = _mm256_add_pd(fjy0,ty);
 918             fjz0             = _mm256_add_pd(fjz0,tz);
 919
 920             }
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 927             {
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq10             = _mm256_mul_pd(iq1,jq0);
 931
 932             /* REACTION-FIELD ELECTROSTATICS */
 933             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 934
 935             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 936
 937             fscal            = felec;
 938
 939             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm256_mul_pd(fscal,dx10);
 943             ty               = _mm256_mul_pd(fscal,dy10);
 944             tz               = _mm256_mul_pd(fscal,dz10);
 945
 946             /* Update vectorial force */
 947             fix1             = _mm256_add_pd(fix1,tx);
 948             fiy1             = _mm256_add_pd(fiy1,ty);
 949             fiz1             = _mm256_add_pd(fiz1,tz);
 950
 951             fjx0             = _mm256_add_pd(fjx0,tx);
 952             fjy0             = _mm256_add_pd(fjy0,ty);
 953             fjz0             = _mm256_add_pd(fjz0,tz);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 962             {
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq20             = _mm256_mul_pd(iq2,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 969
 970             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 975
 976             /* Calculate temporary vectorial force */
 977             tx               = _mm256_mul_pd(fscal,dx20);
 978             ty               = _mm256_mul_pd(fscal,dy20);
 979             tz               = _mm256_mul_pd(fscal,dz20);
 980
 981             /* Update vectorial force */
 982             fix2             = _mm256_add_pd(fix2,tx);
 983             fiy2             = _mm256_add_pd(fiy2,ty);
 984             fiz2             = _mm256_add_pd(fiz2,tz);
 985
 986             fjx0             = _mm256_add_pd(fjx0,tx);
 987             fjy0             = _mm256_add_pd(fjy0,ty);
 988             fjz0             = _mm256_add_pd(fjz0,tz);
 989
 990             }
 991
 992             fjptrA             = f+j_coord_offsetA;
 993             fjptrB             = f+j_coord_offsetB;
 994             fjptrC             = f+j_coord_offsetC;
 995             fjptrD             = f+j_coord_offsetD;
 996
 997             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 998
 999             /* Inner loop uses 124 flops */
1000         }
1001
1002         if(jidx<j_index_end)
1003         {
1004
1005             /* Get j neighbor index, and coordinate index */
1006             jnrlistA         = jjnr[jidx];
1007             jnrlistB         = jjnr[jidx+1];
1008             jnrlistC         = jjnr[jidx+2];
1009             jnrlistD         = jjnr[jidx+3];
1010             /* Sign of each element will be negative for non-real atoms.
1011              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1012              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1013              */
1014             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1015
1016             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1017             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1018             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1019
1020             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1021             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1022             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1023             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1024             j_coord_offsetA  = DIM*jnrA;
1025             j_coord_offsetB  = DIM*jnrB;
1026             j_coord_offsetC  = DIM*jnrC;
1027             j_coord_offsetD  = DIM*jnrD;
1028
1029             /* load j atom coordinates */
1030             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1031                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1032                                                  &jx0,&jy0,&jz0);
1033
1034             /* Calculate displacement vector */
1035             dx00             = _mm256_sub_pd(ix0,jx0);
1036             dy00             = _mm256_sub_pd(iy0,jy0);
1037             dz00             = _mm256_sub_pd(iz0,jz0);
1038             dx10             = _mm256_sub_pd(ix1,jx0);
1039             dy10             = _mm256_sub_pd(iy1,jy0);
1040             dz10             = _mm256_sub_pd(iz1,jz0);
1041             dx20             = _mm256_sub_pd(ix2,jx0);
1042             dy20             = _mm256_sub_pd(iy2,jy0);
1043             dz20             = _mm256_sub_pd(iz2,jz0);
1044
1045             /* Calculate squared distance and things based on it */
1046             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1047             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1048             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1049
1050             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1051             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1052             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1053
1054             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1055             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1056             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1060                                                                  charge+jnrC+0,charge+jnrD+0);
1061             vdwjidx0A        = 2*vdwtype[jnrA+0];
1062             vdwjidx0B        = 2*vdwtype[jnrB+0];
1063             vdwjidx0C        = 2*vdwtype[jnrC+0];
1064             vdwjidx0D        = 2*vdwtype[jnrD+0];
1065
1066             fjx0             = _mm256_setzero_pd();
1067             fjy0             = _mm256_setzero_pd();
1068             fjz0             = _mm256_setzero_pd();
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1075             {
1076
1077             r00              = _mm256_mul_pd(rsq00,rinv00);
1078             r00              = _mm256_andnot_pd(dummy_mask,r00);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq00             = _mm256_mul_pd(iq0,jq0);
1082             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1083                                             vdwioffsetptr0+vdwjidx0B,
1084                                             vdwioffsetptr0+vdwjidx0C,
1085                                             vdwioffsetptr0+vdwjidx0D,
1086                                             &c6_00,&c12_00);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1090
1091             /* LENNARD-JONES DISPERSION/REPULSION */
1092
1093             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1094             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1095             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1096             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1097             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1098
1099             d                = _mm256_sub_pd(r00,rswitch);
1100             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1101             d2               = _mm256_mul_pd(d,d);
1102             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1103
1104             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1105
1106             /* Evaluate switch function */
1107             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1108             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1109             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1110
1111             fscal            = _mm256_add_pd(felec,fvdw);
1112
1113             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1114
1115             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm256_mul_pd(fscal,dx00);
1119             ty               = _mm256_mul_pd(fscal,dy00);
1120             tz               = _mm256_mul_pd(fscal,dz00);
1121
1122             /* Update vectorial force */
1123             fix0             = _mm256_add_pd(fix0,tx);
1124             fiy0             = _mm256_add_pd(fiy0,ty);
1125             fiz0             = _mm256_add_pd(fiz0,tz);
1126
1127             fjx0             = _mm256_add_pd(fjx0,tx);
1128             fjy0             = _mm256_add_pd(fjy0,ty);
1129             fjz0             = _mm256_add_pd(fjz0,tz);
1130
1131             }
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1138             {
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             qq10             = _mm256_mul_pd(iq1,jq0);
1142
1143             /* REACTION-FIELD ELECTROSTATICS */
1144             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1145
1146             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1151
1152             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_pd(fscal,dx10);
1156             ty               = _mm256_mul_pd(fscal,dy10);
1157             tz               = _mm256_mul_pd(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm256_add_pd(fix1,tx);
1161             fiy1             = _mm256_add_pd(fiy1,ty);
1162             fiz1             = _mm256_add_pd(fiz1,tz);
1163
1164             fjx0             = _mm256_add_pd(fjx0,tx);
1165             fjy0             = _mm256_add_pd(fjy0,ty);
1166             fjz0             = _mm256_add_pd(fjz0,tz);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1175             {
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq20             = _mm256_mul_pd(iq2,jq0);
1179
1180             /* REACTION-FIELD ELECTROSTATICS */
1181             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1182
1183             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1184
1185             fscal            = felec;
1186
1187             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1188
1189             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm256_mul_pd(fscal,dx20);
1193             ty               = _mm256_mul_pd(fscal,dy20);
1194             tz               = _mm256_mul_pd(fscal,dz20);
1195
1196             /* Update vectorial force */
1197             fix2             = _mm256_add_pd(fix2,tx);
1198             fiy2             = _mm256_add_pd(fiy2,ty);
1199             fiz2             = _mm256_add_pd(fiz2,tz);
1200
1201             fjx0             = _mm256_add_pd(fjx0,tx);
1202             fjy0             = _mm256_add_pd(fjy0,ty);
1203             fjz0             = _mm256_add_pd(fjz0,tz);
1204
1205             }
1206
1207             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1208             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1209             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1210             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1211
1212             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1213
1214             /* Inner loop uses 125 flops */
1215         }
1216
1217         /* End of innermost loop */
1218
1219         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1220                                                  f+i_coord_offset,fshift+i_shift_offset);
1221
1222         /* Increment number of inner iterations */
1223         inneriter                  += j_index_end - j_index_start;
1224
1225         /* Outer loop uses 18 flops */
1226     }
1227
1228     /* Increment number of outer iterations */
1229     outeriter        += nri;
1230
1231     /* Update outer/inner flops */
1232
1233     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1234 }