src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_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_VdwLJSh_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_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     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m256d          dummy_mask,cutoff_mask;
  93     __m128           tmpmask0,tmpmask1;
  94     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  95     __m256d          one     = _mm256_set1_pd(1.0);
  96     __m256d          two     = _mm256_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm256_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm256_set1_pd(fr->ic->k_rf);
 111     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 112     crf              = _mm256_set1_pd(fr->ic->c_rf);
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[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     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 122     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 128
 129     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 130     rvdw             = _mm256_set1_pd(fr->rvdw);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm256_setzero_pd();
 166         fiy0             = _mm256_setzero_pd();
 167         fiz0             = _mm256_setzero_pd();
 168         fix1             = _mm256_setzero_pd();
 169         fiy1             = _mm256_setzero_pd();
 170         fiz1             = _mm256_setzero_pd();
 171         fix2             = _mm256_setzero_pd();
 172         fiy2             = _mm256_setzero_pd();
 173         fiz2             = _mm256_setzero_pd();
 174         fix3             = _mm256_setzero_pd();
 175         fiy3             = _mm256_setzero_pd();
 176         fiz3             = _mm256_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_pd();
 180         vvdwsum          = _mm256_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 199                                                  &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm256_sub_pd(ix0,jx0);
 203             dy00             = _mm256_sub_pd(iy0,jy0);
 204             dz00             = _mm256_sub_pd(iz0,jz0);
 205             dx10             = _mm256_sub_pd(ix1,jx0);
 206             dy10             = _mm256_sub_pd(iy1,jy0);
 207             dz10             = _mm256_sub_pd(iz1,jz0);
 208             dx20             = _mm256_sub_pd(ix2,jx0);
 209             dy20             = _mm256_sub_pd(iy2,jy0);
 210             dz20             = _mm256_sub_pd(iz2,jz0);
 211             dx30             = _mm256_sub_pd(ix3,jx0);
 212             dy30             = _mm256_sub_pd(iy3,jy0);
 213             dz30             = _mm256_sub_pd(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 217             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 218             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 219             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 220
 221             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 222             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 223             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 224
 225             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 226             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 228             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 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             /* Compute parameters for interactions between i and j atoms */
 250             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 251                                             vdwioffsetptr0+vdwjidx0B,
 252                                             vdwioffsetptr0+vdwjidx0C,
 253                                             vdwioffsetptr0+vdwjidx0D,
 254                                             &c6_00,&c12_00);
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 260             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 261             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 262                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 263             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 264
 265             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 269             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 270
 271             fscal            = fvdw;
 272
 273             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm256_mul_pd(fscal,dx00);
 277             ty               = _mm256_mul_pd(fscal,dy00);
 278             tz               = _mm256_mul_pd(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm256_add_pd(fix0,tx);
 282             fiy0             = _mm256_add_pd(fiy0,ty);
 283             fiz0             = _mm256_add_pd(fiz0,tz);
 284
 285             fjx0             = _mm256_add_pd(fjx0,tx);
 286             fjy0             = _mm256_add_pd(fjy0,ty);
 287             fjz0             = _mm256_add_pd(fjz0,tz);
 288
 289             }
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 296             {
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm256_mul_pd(iq1,jq0);
 300
 301             /* REACTION-FIELD ELECTROSTATICS */
 302             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 303             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 304
 305             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm256_and_pd(velec,cutoff_mask);
 309             velecsum         = _mm256_add_pd(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_pd(fscal,dx10);
 317             ty               = _mm256_mul_pd(fscal,dy10);
 318             tz               = _mm256_mul_pd(fscal,dz10);
 319
 320             /* Update vectorial force */
 321             fix1             = _mm256_add_pd(fix1,tx);
 322             fiy1             = _mm256_add_pd(fiy1,ty);
 323             fiz1             = _mm256_add_pd(fiz1,tz);
 324
 325             fjx0             = _mm256_add_pd(fjx0,tx);
 326             fjy0             = _mm256_add_pd(fjy0,ty);
 327             fjz0             = _mm256_add_pd(fjz0,tz);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm256_mul_pd(iq2,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 343             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 344
 345             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm256_and_pd(velec,cutoff_mask);
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx20);
 357             ty               = _mm256_mul_pd(fscal,dy20);
 358             tz               = _mm256_mul_pd(fscal,dz20);
 359
 360             /* Update vectorial force */
 361             fix2             = _mm256_add_pd(fix2,tx);
 362             fiy2             = _mm256_add_pd(fiy2,ty);
 363             fiz2             = _mm256_add_pd(fiz2,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq30             = _mm256_mul_pd(iq3,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 383             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 384
 385             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm256_and_pd(velec,cutoff_mask);
 389             velecsum         = _mm256_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_pd(fscal,dx30);
 397             ty               = _mm256_mul_pd(fscal,dy30);
 398             tz               = _mm256_mul_pd(fscal,dz30);
 399
 400             /* Update vectorial force */
 401             fix3             = _mm256_add_pd(fix3,tx);
 402             fiy3             = _mm256_add_pd(fiy3,ty);
 403             fiz3             = _mm256_add_pd(fiz3,tz);
 404
 405             fjx0             = _mm256_add_pd(fjx0,tx);
 406             fjy0             = _mm256_add_pd(fjy0,ty);
 407             fjz0             = _mm256_add_pd(fjz0,tz);
 408
 409             }
 410
 411             fjptrA             = f+j_coord_offsetA;
 412             fjptrB             = f+j_coord_offsetB;
 413             fjptrC             = f+j_coord_offsetC;
 414             fjptrD             = f+j_coord_offsetD;
 415
 416             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 152 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 432              */
 433             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434
 435             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 436             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 437             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 438
 439             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 440             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 441             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 442             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 443             j_coord_offsetA  = DIM*jnrA;
 444             j_coord_offsetB  = DIM*jnrB;
 445             j_coord_offsetC  = DIM*jnrC;
 446             j_coord_offsetD  = DIM*jnrD;
 447
 448             /* load j atom coordinates */
 449             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 450                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 451                                                  &jx0,&jy0,&jz0);
 452
 453             /* Calculate displacement vector */
 454             dx00             = _mm256_sub_pd(ix0,jx0);
 455             dy00             = _mm256_sub_pd(iy0,jy0);
 456             dz00             = _mm256_sub_pd(iz0,jz0);
 457             dx10             = _mm256_sub_pd(ix1,jx0);
 458             dy10             = _mm256_sub_pd(iy1,jy0);
 459             dz10             = _mm256_sub_pd(iz1,jz0);
 460             dx20             = _mm256_sub_pd(ix2,jx0);
 461             dy20             = _mm256_sub_pd(iy2,jy0);
 462             dz20             = _mm256_sub_pd(iz2,jz0);
 463             dx30             = _mm256_sub_pd(ix3,jx0);
 464             dy30             = _mm256_sub_pd(iy3,jy0);
 465             dz30             = _mm256_sub_pd(iz3,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 469             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 470             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 471             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 472
 473             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 474             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 475             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 476
 477             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 478             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 479             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 480             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487             vdwjidx0C        = 2*vdwtype[jnrC+0];
 488             vdwjidx0D        = 2*vdwtype[jnrD+0];
 489
 490             fjx0             = _mm256_setzero_pd();
 491             fjy0             = _mm256_setzero_pd();
 492             fjz0             = _mm256_setzero_pd();
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 499             {
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 503                                             vdwioffsetptr0+vdwjidx0B,
 504                                             vdwioffsetptr0+vdwjidx0C,
 505                                             vdwioffsetptr0+vdwjidx0D,
 506                                             &c6_00,&c12_00);
 507
 508             /* LENNARD-JONES DISPERSION/REPULSION */
 509
 510             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 511             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 512             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 513             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 514                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 515             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 516
 517             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 521             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 522             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 527
 528             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm256_mul_pd(fscal,dx00);
 532             ty               = _mm256_mul_pd(fscal,dy00);
 533             tz               = _mm256_mul_pd(fscal,dz00);
 534
 535             /* Update vectorial force */
 536             fix0             = _mm256_add_pd(fix0,tx);
 537             fiy0             = _mm256_add_pd(fiy0,ty);
 538             fiz0             = _mm256_add_pd(fiz0,tz);
 539
 540             fjx0             = _mm256_add_pd(fjx0,tx);
 541             fjy0             = _mm256_add_pd(fjy0,ty);
 542             fjz0             = _mm256_add_pd(fjz0,tz);
 543
 544             }
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 551             {
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq10             = _mm256_mul_pd(iq1,jq0);
 555
 556             /* REACTION-FIELD ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 558             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 559
 560             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm256_and_pd(velec,cutoff_mask);
 564             velec            = _mm256_andnot_pd(dummy_mask,velec);
 565             velecsum         = _mm256_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 570
 571             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx10);
 575             ty               = _mm256_mul_pd(fscal,dy10);
 576             tz               = _mm256_mul_pd(fscal,dz10);
 577
 578             /* Update vectorial force */
 579             fix1             = _mm256_add_pd(fix1,tx);
 580             fiy1             = _mm256_add_pd(fiy1,ty);
 581             fiz1             = _mm256_add_pd(fiz1,tz);
 582
 583             fjx0             = _mm256_add_pd(fjx0,tx);
 584             fjy0             = _mm256_add_pd(fjy0,ty);
 585             fjz0             = _mm256_add_pd(fjz0,tz);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 594             {
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq20             = _mm256_mul_pd(iq2,jq0);
 598
 599             /* REACTION-FIELD ELECTROSTATICS */
 600             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 601             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 602
 603             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm256_and_pd(velec,cutoff_mask);
 607             velec            = _mm256_andnot_pd(dummy_mask,velec);
 608             velecsum         = _mm256_add_pd(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 613
 614             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm256_mul_pd(fscal,dx20);
 618             ty               = _mm256_mul_pd(fscal,dy20);
 619             tz               = _mm256_mul_pd(fscal,dz20);
 620
 621             /* Update vectorial force */
 622             fix2             = _mm256_add_pd(fix2,tx);
 623             fiy2             = _mm256_add_pd(fiy2,ty);
 624             fiz2             = _mm256_add_pd(fiz2,tz);
 625
 626             fjx0             = _mm256_add_pd(fjx0,tx);
 627             fjy0             = _mm256_add_pd(fjy0,ty);
 628             fjz0             = _mm256_add_pd(fjz0,tz);
 629
 630             }
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 637             {
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             qq30             = _mm256_mul_pd(iq3,jq0);
 641
 642             /* REACTION-FIELD ELECTROSTATICS */
 643             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 644             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 645
 646             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm256_and_pd(velec,cutoff_mask);
 650             velec            = _mm256_andnot_pd(dummy_mask,velec);
 651             velecsum         = _mm256_add_pd(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 656
 657             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_pd(fscal,dx30);
 661             ty               = _mm256_mul_pd(fscal,dy30);
 662             tz               = _mm256_mul_pd(fscal,dz30);
 663
 664             /* Update vectorial force */
 665             fix3             = _mm256_add_pd(fix3,tx);
 666             fiy3             = _mm256_add_pd(fiy3,ty);
 667             fiz3             = _mm256_add_pd(fiz3,tz);
 668
 669             fjx0             = _mm256_add_pd(fjx0,tx);
 670             fjy0             = _mm256_add_pd(fjy0,ty);
 671             fjz0             = _mm256_add_pd(fjz0,tz);
 672
 673             }
 674
 675             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 676             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 677             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 678             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 679
 680             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 681
 682             /* Inner loop uses 152 flops */
 683         }
 684
 685         /* End of innermost loop */
 686
 687         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 688                                                  f+i_coord_offset,fshift+i_shift_offset);
 689
 690         ggid                        = gid[iidx];
 691         /* Update potential energies */
 692         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 693         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 694
 695         /* Increment number of inner iterations */
 696         inneriter                  += j_index_end - j_index_start;
 697
 698         /* Outer loop uses 26 flops */
 699     }
 700
 701     /* Increment number of outer iterations */
 702     outeriter        += nri;
 703
 704     /* Update outer/inner flops */
 705
 706     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 707 }
 708 /*
 709  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 710  * Electrostatics interaction: ReactionField
 711  * VdW interaction:            LennardJones
 712  * Geometry:                   Water4-Particle
 713  * Calculate force/pot:        Force
 714  */
 715 void
 716 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 717                     (t_nblist * gmx_restrict                nlist,
 718                      rvec * gmx_restrict                    xx,
 719                      rvec * gmx_restrict                    ff,
 720                      t_forcerec * gmx_restrict              fr,
 721                      t_mdatoms * gmx_restrict               mdatoms,
 722                      nb_kernel_data_t * gmx_restrict        kernel_data,
 723                      t_nrnb * gmx_restrict                  nrnb)
 724 {
 725     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 726      * just 0 for non-waters.
 727      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 728      * jnr indices corresponding to data put in the four positions in the SIMD register.
 729      */
 730     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 731     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 732     int              jnrA,jnrB,jnrC,jnrD;
 733     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 734     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 735     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 736     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 737     real             rcutoff_scalar;
 738     real             *shiftvec,*fshift,*x,*f;
 739     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 740     real             scratch[4*DIM];
 741     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 742     real *           vdwioffsetptr0;
 743     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 744     real *           vdwioffsetptr1;
 745     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 746     real *           vdwioffsetptr2;
 747     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 748     real *           vdwioffsetptr3;
 749     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 750     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 751     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 752     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 753     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 754     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 755     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 756     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 757     real             *charge;
 758     int              nvdwtype;
 759     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 760     int              *vdwtype;
 761     real             *vdwparam;
 762     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 763     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 764     __m256d          dummy_mask,cutoff_mask;
 765     __m128           tmpmask0,tmpmask1;
 766     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 767     __m256d          one     = _mm256_set1_pd(1.0);
 768     __m256d          two     = _mm256_set1_pd(2.0);
 769     x                = xx[0];
 770     f                = ff[0];
 771
 772     nri              = nlist->nri;
 773     iinr             = nlist->iinr;
 774     jindex           = nlist->jindex;
 775     jjnr             = nlist->jjnr;
 776     shiftidx         = nlist->shift;
 777     gid              = nlist->gid;
 778     shiftvec         = fr->shift_vec[0];
 779     fshift           = fr->fshift[0];
 780     facel            = _mm256_set1_pd(fr->epsfac);
 781     charge           = mdatoms->chargeA;
 782     krf              = _mm256_set1_pd(fr->ic->k_rf);
 783     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 784     crf              = _mm256_set1_pd(fr->ic->c_rf);
 785     nvdwtype         = fr->ntype;
 786     vdwparam         = fr->nbfp;
 787     vdwtype          = mdatoms->typeA;
 788
 789     /* Setup water-specific parameters */
 790     inr              = nlist->iinr[0];
 791     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 792     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 793     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 794     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 795
 796     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 797     rcutoff_scalar   = fr->rcoulomb;
 798     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 799     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 800
 801     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 802     rvdw             = _mm256_set1_pd(fr->rvdw);
 803
 804     /* Avoid stupid compiler warnings */
 805     jnrA = jnrB = jnrC = jnrD = 0;
 806     j_coord_offsetA = 0;
 807     j_coord_offsetB = 0;
 808     j_coord_offsetC = 0;
 809     j_coord_offsetD = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     for(iidx=0;iidx<4*DIM;iidx++)
 815     {
 816         scratch[iidx] = 0.0;
 817     }
 818
 819     /* Start outer loop over neighborlists */
 820     for(iidx=0; iidx<nri; iidx++)
 821     {
 822         /* Load shift vector for this list */
 823         i_shift_offset   = DIM*shiftidx[iidx];
 824
 825         /* Load limits for loop over neighbors */
 826         j_index_start    = jindex[iidx];
 827         j_index_end      = jindex[iidx+1];
 828
 829         /* Get outer coordinate index */
 830         inr              = iinr[iidx];
 831         i_coord_offset   = DIM*inr;
 832
 833         /* Load i particle coords and add shift vector */
 834         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 835                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 836
 837         fix0             = _mm256_setzero_pd();
 838         fiy0             = _mm256_setzero_pd();
 839         fiz0             = _mm256_setzero_pd();
 840         fix1             = _mm256_setzero_pd();
 841         fiy1             = _mm256_setzero_pd();
 842         fiz1             = _mm256_setzero_pd();
 843         fix2             = _mm256_setzero_pd();
 844         fiy2             = _mm256_setzero_pd();
 845         fiz2             = _mm256_setzero_pd();
 846         fix3             = _mm256_setzero_pd();
 847         fiy3             = _mm256_setzero_pd();
 848         fiz3             = _mm256_setzero_pd();
 849
 850         /* Start inner kernel loop */
 851         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 852         {
 853
 854             /* Get j neighbor index, and coordinate index */
 855             jnrA             = jjnr[jidx];
 856             jnrB             = jjnr[jidx+1];
 857             jnrC             = jjnr[jidx+2];
 858             jnrD             = jjnr[jidx+3];
 859             j_coord_offsetA  = DIM*jnrA;
 860             j_coord_offsetB  = DIM*jnrB;
 861             j_coord_offsetC  = DIM*jnrC;
 862             j_coord_offsetD  = DIM*jnrD;
 863
 864             /* load j atom coordinates */
 865             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 866                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 867                                                  &jx0,&jy0,&jz0);
 868
 869             /* Calculate displacement vector */
 870             dx00             = _mm256_sub_pd(ix0,jx0);
 871             dy00             = _mm256_sub_pd(iy0,jy0);
 872             dz00             = _mm256_sub_pd(iz0,jz0);
 873             dx10             = _mm256_sub_pd(ix1,jx0);
 874             dy10             = _mm256_sub_pd(iy1,jy0);
 875             dz10             = _mm256_sub_pd(iz1,jz0);
 876             dx20             = _mm256_sub_pd(ix2,jx0);
 877             dy20             = _mm256_sub_pd(iy2,jy0);
 878             dz20             = _mm256_sub_pd(iz2,jz0);
 879             dx30             = _mm256_sub_pd(ix3,jx0);
 880             dy30             = _mm256_sub_pd(iy3,jy0);
 881             dz30             = _mm256_sub_pd(iz3,jz0);
 882
 883             /* Calculate squared distance and things based on it */
 884             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 885             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 886             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 887             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 888
 889             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 890             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 891             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 892
 893             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 894             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 895             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 896             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 897
 898             /* Load parameters for j particles */
 899             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 900                                                                  charge+jnrC+0,charge+jnrD+0);
 901             vdwjidx0A        = 2*vdwtype[jnrA+0];
 902             vdwjidx0B        = 2*vdwtype[jnrB+0];
 903             vdwjidx0C        = 2*vdwtype[jnrC+0];
 904             vdwjidx0D        = 2*vdwtype[jnrD+0];
 905
 906             fjx0             = _mm256_setzero_pd();
 907             fjy0             = _mm256_setzero_pd();
 908             fjz0             = _mm256_setzero_pd();
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 915             {
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 919                                             vdwioffsetptr0+vdwjidx0B,
 920                                             vdwioffsetptr0+vdwjidx0C,
 921                                             vdwioffsetptr0+vdwjidx0D,
 922                                             &c6_00,&c12_00);
 923
 924             /* LENNARD-JONES DISPERSION/REPULSION */
 925
 926             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 927             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 928
 929             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 930
 931             fscal            = fvdw;
 932
 933             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm256_mul_pd(fscal,dx00);
 937             ty               = _mm256_mul_pd(fscal,dy00);
 938             tz               = _mm256_mul_pd(fscal,dz00);
 939
 940             /* Update vectorial force */
 941             fix0             = _mm256_add_pd(fix0,tx);
 942             fiy0             = _mm256_add_pd(fiy0,ty);
 943             fiz0             = _mm256_add_pd(fiz0,tz);
 944
 945             fjx0             = _mm256_add_pd(fjx0,tx);
 946             fjy0             = _mm256_add_pd(fjy0,ty);
 947             fjz0             = _mm256_add_pd(fjz0,tz);
 948
 949             }
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 956             {
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq10             = _mm256_mul_pd(iq1,jq0);
 960
 961             /* REACTION-FIELD ELECTROSTATICS */
 962             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 963
 964             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 965
 966             fscal            = felec;
 967
 968             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm256_mul_pd(fscal,dx10);
 972             ty               = _mm256_mul_pd(fscal,dy10);
 973             tz               = _mm256_mul_pd(fscal,dz10);
 974
 975             /* Update vectorial force */
 976             fix1             = _mm256_add_pd(fix1,tx);
 977             fiy1             = _mm256_add_pd(fiy1,ty);
 978             fiz1             = _mm256_add_pd(fiz1,tz);
 979
 980             fjx0             = _mm256_add_pd(fjx0,tx);
 981             fjy0             = _mm256_add_pd(fjy0,ty);
 982             fjz0             = _mm256_add_pd(fjz0,tz);
 983
 984             }
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 991             {
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq20             = _mm256_mul_pd(iq2,jq0);
 995
 996             /* REACTION-FIELD ELECTROSTATICS */
 997             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 998
 999             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm256_mul_pd(fscal,dx20);
1007             ty               = _mm256_mul_pd(fscal,dy20);
1008             tz               = _mm256_mul_pd(fscal,dz20);
1009
1010             /* Update vectorial force */
1011             fix2             = _mm256_add_pd(fix2,tx);
1012             fiy2             = _mm256_add_pd(fiy2,ty);
1013             fiz2             = _mm256_add_pd(fiz2,tz);
1014
1015             fjx0             = _mm256_add_pd(fjx0,tx);
1016             fjy0             = _mm256_add_pd(fjy0,ty);
1017             fjz0             = _mm256_add_pd(fjz0,tz);
1018
1019             }
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1026             {
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq30             = _mm256_mul_pd(iq3,jq0);
1030
1031             /* REACTION-FIELD ELECTROSTATICS */
1032             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1033
1034             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1035
1036             fscal            = felec;
1037
1038             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_pd(fscal,dx30);
1042             ty               = _mm256_mul_pd(fscal,dy30);
1043             tz               = _mm256_mul_pd(fscal,dz30);
1044
1045             /* Update vectorial force */
1046             fix3             = _mm256_add_pd(fix3,tx);
1047             fiy3             = _mm256_add_pd(fiy3,ty);
1048             fiz3             = _mm256_add_pd(fiz3,tz);
1049
1050             fjx0             = _mm256_add_pd(fjx0,tx);
1051             fjy0             = _mm256_add_pd(fjy0,ty);
1052             fjz0             = _mm256_add_pd(fjz0,tz);
1053
1054             }
1055
1056             fjptrA             = f+j_coord_offsetA;
1057             fjptrB             = f+j_coord_offsetB;
1058             fjptrC             = f+j_coord_offsetC;
1059             fjptrD             = f+j_coord_offsetD;
1060
1061             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1062
1063             /* Inner loop uses 123 flops */
1064         }
1065
1066         if(jidx<j_index_end)
1067         {
1068
1069             /* Get j neighbor index, and coordinate index */
1070             jnrlistA         = jjnr[jidx];
1071             jnrlistB         = jjnr[jidx+1];
1072             jnrlistC         = jjnr[jidx+2];
1073             jnrlistD         = jjnr[jidx+3];
1074             /* Sign of each element will be negative for non-real atoms.
1075              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1076              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1077              */
1078             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1079
1080             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1081             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1082             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1083
1084             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1085             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1086             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1087             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1088             j_coord_offsetA  = DIM*jnrA;
1089             j_coord_offsetB  = DIM*jnrB;
1090             j_coord_offsetC  = DIM*jnrC;
1091             j_coord_offsetD  = DIM*jnrD;
1092
1093             /* load j atom coordinates */
1094             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1095                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1096                                                  &jx0,&jy0,&jz0);
1097
1098             /* Calculate displacement vector */
1099             dx00             = _mm256_sub_pd(ix0,jx0);
1100             dy00             = _mm256_sub_pd(iy0,jy0);
1101             dz00             = _mm256_sub_pd(iz0,jz0);
1102             dx10             = _mm256_sub_pd(ix1,jx0);
1103             dy10             = _mm256_sub_pd(iy1,jy0);
1104             dz10             = _mm256_sub_pd(iz1,jz0);
1105             dx20             = _mm256_sub_pd(ix2,jx0);
1106             dy20             = _mm256_sub_pd(iy2,jy0);
1107             dz20             = _mm256_sub_pd(iz2,jz0);
1108             dx30             = _mm256_sub_pd(ix3,jx0);
1109             dy30             = _mm256_sub_pd(iy3,jy0);
1110             dz30             = _mm256_sub_pd(iz3,jz0);
1111
1112             /* Calculate squared distance and things based on it */
1113             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1114             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1115             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1116             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1117
1118             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1119             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1120             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1121
1122             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1123             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1124             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1125             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1126
1127             /* Load parameters for j particles */
1128             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1129                                                                  charge+jnrC+0,charge+jnrD+0);
1130             vdwjidx0A        = 2*vdwtype[jnrA+0];
1131             vdwjidx0B        = 2*vdwtype[jnrB+0];
1132             vdwjidx0C        = 2*vdwtype[jnrC+0];
1133             vdwjidx0D        = 2*vdwtype[jnrD+0];
1134
1135             fjx0             = _mm256_setzero_pd();
1136             fjy0             = _mm256_setzero_pd();
1137             fjz0             = _mm256_setzero_pd();
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1144             {
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1148                                             vdwioffsetptr0+vdwjidx0B,
1149                                             vdwioffsetptr0+vdwjidx0C,
1150                                             vdwioffsetptr0+vdwjidx0D,
1151                                             &c6_00,&c12_00);
1152
1153             /* LENNARD-JONES DISPERSION/REPULSION */
1154
1155             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1156             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1157
1158             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1159
1160             fscal            = fvdw;
1161
1162             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1163
1164             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm256_mul_pd(fscal,dx00);
1168             ty               = _mm256_mul_pd(fscal,dy00);
1169             tz               = _mm256_mul_pd(fscal,dz00);
1170
1171             /* Update vectorial force */
1172             fix0             = _mm256_add_pd(fix0,tx);
1173             fiy0             = _mm256_add_pd(fiy0,ty);
1174             fiz0             = _mm256_add_pd(fiz0,tz);
1175
1176             fjx0             = _mm256_add_pd(fjx0,tx);
1177             fjy0             = _mm256_add_pd(fjy0,ty);
1178             fjz0             = _mm256_add_pd(fjz0,tz);
1179
1180             }
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1187             {
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq10             = _mm256_mul_pd(iq1,jq0);
1191
1192             /* REACTION-FIELD ELECTROSTATICS */
1193             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1194
1195             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1200
1201             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm256_mul_pd(fscal,dx10);
1205             ty               = _mm256_mul_pd(fscal,dy10);
1206             tz               = _mm256_mul_pd(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm256_add_pd(fix1,tx);
1210             fiy1             = _mm256_add_pd(fiy1,ty);
1211             fiz1             = _mm256_add_pd(fiz1,tz);
1212
1213             fjx0             = _mm256_add_pd(fjx0,tx);
1214             fjy0             = _mm256_add_pd(fjy0,ty);
1215             fjz0             = _mm256_add_pd(fjz0,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1224             {
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq20             = _mm256_mul_pd(iq2,jq0);
1228
1229             /* REACTION-FIELD ELECTROSTATICS */
1230             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1231
1232             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1233
1234             fscal            = felec;
1235
1236             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1237
1238             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1239
1240             /* Calculate temporary vectorial force */
1241             tx               = _mm256_mul_pd(fscal,dx20);
1242             ty               = _mm256_mul_pd(fscal,dy20);
1243             tz               = _mm256_mul_pd(fscal,dz20);
1244
1245             /* Update vectorial force */
1246             fix2             = _mm256_add_pd(fix2,tx);
1247             fiy2             = _mm256_add_pd(fiy2,ty);
1248             fiz2             = _mm256_add_pd(fiz2,tz);
1249
1250             fjx0             = _mm256_add_pd(fjx0,tx);
1251             fjy0             = _mm256_add_pd(fjy0,ty);
1252             fjz0             = _mm256_add_pd(fjz0,tz);
1253
1254             }
1255
1256             /**************************
1257              * CALCULATE INTERACTIONS *
1258              **************************/
1259
1260             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1261             {
1262
1263             /* Compute parameters for interactions between i and j atoms */
1264             qq30             = _mm256_mul_pd(iq3,jq0);
1265
1266             /* REACTION-FIELD ELECTROSTATICS */
1267             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1268
1269             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1270
1271             fscal            = felec;
1272
1273             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1274
1275             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1276
1277             /* Calculate temporary vectorial force */
1278             tx               = _mm256_mul_pd(fscal,dx30);
1279             ty               = _mm256_mul_pd(fscal,dy30);
1280             tz               = _mm256_mul_pd(fscal,dz30);
1281
1282             /* Update vectorial force */
1283             fix3             = _mm256_add_pd(fix3,tx);
1284             fiy3             = _mm256_add_pd(fiy3,ty);
1285             fiz3             = _mm256_add_pd(fiz3,tz);
1286
1287             fjx0             = _mm256_add_pd(fjx0,tx);
1288             fjy0             = _mm256_add_pd(fjy0,ty);
1289             fjz0             = _mm256_add_pd(fjz0,tz);
1290
1291             }
1292
1293             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1294             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1295             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1296             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1297
1298             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1299
1300             /* Inner loop uses 123 flops */
1301         }
1302
1303         /* End of innermost loop */
1304
1305         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1306                                                  f+i_coord_offset,fshift+i_shift_offset);
1307
1308         /* Increment number of inner iterations */
1309         inneriter                  += j_index_end - j_index_start;
1310
1311         /* Outer loop uses 24 flops */
1312     }
1313
1314     /* Increment number of outer iterations */
1315     outeriter        += nri;
1316
1317     /* Update outer/inner flops */
1318
1319     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1320 }