src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_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     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 117     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 118     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 119     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 152                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 153
 154         fix0             = _mm256_setzero_pd();
 155         fiy0             = _mm256_setzero_pd();
 156         fiz0             = _mm256_setzero_pd();
 157         fix1             = _mm256_setzero_pd();
 158         fiy1             = _mm256_setzero_pd();
 159         fiz1             = _mm256_setzero_pd();
 160         fix2             = _mm256_setzero_pd();
 161         fiy2             = _mm256_setzero_pd();
 162         fiz2             = _mm256_setzero_pd();
 163         fix3             = _mm256_setzero_pd();
 164         fiy3             = _mm256_setzero_pd();
 165         fiz3             = _mm256_setzero_pd();
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm256_setzero_pd();
 169         vvdwsum          = _mm256_setzero_pd();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 188                                                  &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm256_sub_pd(ix0,jx0);
 192             dy00             = _mm256_sub_pd(iy0,jy0);
 193             dz00             = _mm256_sub_pd(iz0,jz0);
 194             dx10             = _mm256_sub_pd(ix1,jx0);
 195             dy10             = _mm256_sub_pd(iy1,jy0);
 196             dz10             = _mm256_sub_pd(iz1,jz0);
 197             dx20             = _mm256_sub_pd(ix2,jx0);
 198             dy20             = _mm256_sub_pd(iy2,jy0);
 199             dz20             = _mm256_sub_pd(iz2,jz0);
 200             dx30             = _mm256_sub_pd(ix3,jx0);
 201             dy30             = _mm256_sub_pd(iy3,jy0);
 202             dz30             = _mm256_sub_pd(iz3,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 206             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 207             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 208             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 209
 210             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 211             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 212             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 213
 214             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 215             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 216             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 217             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             fjx0             = _mm256_setzero_pd();
 228             fjy0             = _mm256_setzero_pd();
 229             fjz0             = _mm256_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 237                                             vdwioffsetptr0+vdwjidx0B,
 238                                             vdwioffsetptr0+vdwjidx0C,
 239                                             vdwioffsetptr0+vdwjidx0D,
 240                                             &c6_00,&c12_00);
 241
 242             /* LENNARD-JONES DISPERSION/REPULSION */
 243
 244             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 245             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 246             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 247             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 248             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 252
 253             fscal            = fvdw;
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm256_mul_pd(fscal,dx00);
 257             ty               = _mm256_mul_pd(fscal,dy00);
 258             tz               = _mm256_mul_pd(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm256_add_pd(fix0,tx);
 262             fiy0             = _mm256_add_pd(fiy0,ty);
 263             fiz0             = _mm256_add_pd(fiz0,tz);
 264
 265             fjx0             = _mm256_add_pd(fjx0,tx);
 266             fjy0             = _mm256_add_pd(fjy0,ty);
 267             fjz0             = _mm256_add_pd(fjz0,tz);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             qq10             = _mm256_mul_pd(iq1,jq0);
 275
 276             /* COULOMB ELECTROSTATICS */
 277             velec            = _mm256_mul_pd(qq10,rinv10);
 278             felec            = _mm256_mul_pd(velec,rinvsq10);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_pd(velecsum,velec);
 282
 283             fscal            = felec;
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_pd(fscal,dx10);
 287             ty               = _mm256_mul_pd(fscal,dy10);
 288             tz               = _mm256_mul_pd(fscal,dz10);
 289
 290             /* Update vectorial force */
 291             fix1             = _mm256_add_pd(fix1,tx);
 292             fiy1             = _mm256_add_pd(fiy1,ty);
 293             fiz1             = _mm256_add_pd(fiz1,tz);
 294
 295             fjx0             = _mm256_add_pd(fjx0,tx);
 296             fjy0             = _mm256_add_pd(fjy0,ty);
 297             fjz0             = _mm256_add_pd(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq20             = _mm256_mul_pd(iq2,jq0);
 305
 306             /* COULOMB ELECTROSTATICS */
 307             velec            = _mm256_mul_pd(qq20,rinv20);
 308             felec            = _mm256_mul_pd(velec,rinvsq20);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm256_add_pd(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_pd(fscal,dx20);
 317             ty               = _mm256_mul_pd(fscal,dy20);
 318             tz               = _mm256_mul_pd(fscal,dz20);
 319
 320             /* Update vectorial force */
 321             fix2             = _mm256_add_pd(fix2,tx);
 322             fiy2             = _mm256_add_pd(fiy2,ty);
 323             fiz2             = _mm256_add_pd(fiz2,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              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq30             = _mm256_mul_pd(iq3,jq0);
 335
 336             /* COULOMB ELECTROSTATICS */
 337             velec            = _mm256_mul_pd(qq30,rinv30);
 338             felec            = _mm256_mul_pd(velec,rinvsq30);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm256_add_pd(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm256_mul_pd(fscal,dx30);
 347             ty               = _mm256_mul_pd(fscal,dy30);
 348             tz               = _mm256_mul_pd(fscal,dz30);
 349
 350             /* Update vectorial force */
 351             fix3             = _mm256_add_pd(fix3,tx);
 352             fiy3             = _mm256_add_pd(fiy3,ty);
 353             fiz3             = _mm256_add_pd(fiz3,tz);
 354
 355             fjx0             = _mm256_add_pd(fjx0,tx);
 356             fjy0             = _mm256_add_pd(fjy0,ty);
 357             fjz0             = _mm256_add_pd(fjz0,tz);
 358
 359             fjptrA             = f+j_coord_offsetA;
 360             fjptrB             = f+j_coord_offsetB;
 361             fjptrC             = f+j_coord_offsetC;
 362             fjptrD             = f+j_coord_offsetD;
 363
 364             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 365
 366             /* Inner loop uses 116 flops */
 367         }
 368
 369         if(jidx<j_index_end)
 370         {
 371
 372             /* Get j neighbor index, and coordinate index */
 373             jnrlistA         = jjnr[jidx];
 374             jnrlistB         = jjnr[jidx+1];
 375             jnrlistC         = jjnr[jidx+2];
 376             jnrlistD         = jjnr[jidx+3];
 377             /* Sign of each element will be negative for non-real atoms.
 378              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 379              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 380              */
 381             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 382
 383             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 384             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 385             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 386
 387             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 388             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 389             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 390             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 391             j_coord_offsetA  = DIM*jnrA;
 392             j_coord_offsetB  = DIM*jnrB;
 393             j_coord_offsetC  = DIM*jnrC;
 394             j_coord_offsetD  = DIM*jnrD;
 395
 396             /* load j atom coordinates */
 397             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 398                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 399                                                  &jx0,&jy0,&jz0);
 400
 401             /* Calculate displacement vector */
 402             dx00             = _mm256_sub_pd(ix0,jx0);
 403             dy00             = _mm256_sub_pd(iy0,jy0);
 404             dz00             = _mm256_sub_pd(iz0,jz0);
 405             dx10             = _mm256_sub_pd(ix1,jx0);
 406             dy10             = _mm256_sub_pd(iy1,jy0);
 407             dz10             = _mm256_sub_pd(iz1,jz0);
 408             dx20             = _mm256_sub_pd(ix2,jx0);
 409             dy20             = _mm256_sub_pd(iy2,jy0);
 410             dz20             = _mm256_sub_pd(iz2,jz0);
 411             dx30             = _mm256_sub_pd(ix3,jx0);
 412             dy30             = _mm256_sub_pd(iy3,jy0);
 413             dz30             = _mm256_sub_pd(iz3,jz0);
 414
 415             /* Calculate squared distance and things based on it */
 416             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 417             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 418             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 419             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 420
 421             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 422             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 423             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 424
 425             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 426             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 427             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 428             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 429
 430             /* Load parameters for j particles */
 431             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 432                                                                  charge+jnrC+0,charge+jnrD+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434             vdwjidx0B        = 2*vdwtype[jnrB+0];
 435             vdwjidx0C        = 2*vdwtype[jnrC+0];
 436             vdwjidx0D        = 2*vdwtype[jnrD+0];
 437
 438             fjx0             = _mm256_setzero_pd();
 439             fjy0             = _mm256_setzero_pd();
 440             fjz0             = _mm256_setzero_pd();
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 448                                             vdwioffsetptr0+vdwjidx0B,
 449                                             vdwioffsetptr0+vdwjidx0C,
 450                                             vdwioffsetptr0+vdwjidx0D,
 451                                             &c6_00,&c12_00);
 452
 453             /* LENNARD-JONES DISPERSION/REPULSION */
 454
 455             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 456             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 457             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 458             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 459             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 463             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 464
 465             fscal            = fvdw;
 466
 467             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_pd(fscal,dx00);
 471             ty               = _mm256_mul_pd(fscal,dy00);
 472             tz               = _mm256_mul_pd(fscal,dz00);
 473
 474             /* Update vectorial force */
 475             fix0             = _mm256_add_pd(fix0,tx);
 476             fiy0             = _mm256_add_pd(fiy0,ty);
 477             fiz0             = _mm256_add_pd(fiz0,tz);
 478
 479             fjx0             = _mm256_add_pd(fjx0,tx);
 480             fjy0             = _mm256_add_pd(fjy0,ty);
 481             fjz0             = _mm256_add_pd(fjz0,tz);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq10             = _mm256_mul_pd(iq1,jq0);
 489
 490             /* COULOMB ELECTROSTATICS */
 491             velec            = _mm256_mul_pd(qq10,rinv10);
 492             felec            = _mm256_mul_pd(velec,rinvsq10);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_andnot_pd(dummy_mask,velec);
 496             velecsum         = _mm256_add_pd(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_pd(fscal,dx10);
 504             ty               = _mm256_mul_pd(fscal,dy10);
 505             tz               = _mm256_mul_pd(fscal,dz10);
 506
 507             /* Update vectorial force */
 508             fix1             = _mm256_add_pd(fix1,tx);
 509             fiy1             = _mm256_add_pd(fiy1,ty);
 510             fiz1             = _mm256_add_pd(fiz1,tz);
 511
 512             fjx0             = _mm256_add_pd(fjx0,tx);
 513             fjy0             = _mm256_add_pd(fjy0,ty);
 514             fjz0             = _mm256_add_pd(fjz0,tz);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq20             = _mm256_mul_pd(iq2,jq0);
 522
 523             /* COULOMB ELECTROSTATICS */
 524             velec            = _mm256_mul_pd(qq20,rinv20);
 525             felec            = _mm256_mul_pd(velec,rinvsq20);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm256_andnot_pd(dummy_mask,velec);
 529             velecsum         = _mm256_add_pd(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_pd(fscal,dx20);
 537             ty               = _mm256_mul_pd(fscal,dy20);
 538             tz               = _mm256_mul_pd(fscal,dz20);
 539
 540             /* Update vectorial force */
 541             fix2             = _mm256_add_pd(fix2,tx);
 542             fiy2             = _mm256_add_pd(fiy2,ty);
 543             fiz2             = _mm256_add_pd(fiz2,tz);
 544
 545             fjx0             = _mm256_add_pd(fjx0,tx);
 546             fjy0             = _mm256_add_pd(fjy0,ty);
 547             fjz0             = _mm256_add_pd(fjz0,tz);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq30             = _mm256_mul_pd(iq3,jq0);
 555
 556             /* COULOMB ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq30,rinv30);
 558             felec            = _mm256_mul_pd(velec,rinvsq30);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm256_andnot_pd(dummy_mask,velec);
 562             velecsum         = _mm256_add_pd(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_pd(fscal,dx30);
 570             ty               = _mm256_mul_pd(fscal,dy30);
 571             tz               = _mm256_mul_pd(fscal,dz30);
 572
 573             /* Update vectorial force */
 574             fix3             = _mm256_add_pd(fix3,tx);
 575             fiy3             = _mm256_add_pd(fiy3,ty);
 576             fiz3             = _mm256_add_pd(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 583             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 584             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 585             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 586
 587             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 588
 589             /* Inner loop uses 116 flops */
 590         }
 591
 592         /* End of innermost loop */
 593
 594         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 595                                                  f+i_coord_offset,fshift+i_shift_offset);
 596
 597         ggid                        = gid[iidx];
 598         /* Update potential energies */
 599         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 600         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 601
 602         /* Increment number of inner iterations */
 603         inneriter                  += j_index_end - j_index_start;
 604
 605         /* Outer loop uses 26 flops */
 606     }
 607
 608     /* Increment number of outer iterations */
 609     outeriter        += nri;
 610
 611     /* Update outer/inner flops */
 612
 613     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 614 }
 615 /*
 616  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 617  * Electrostatics interaction: Coulomb
 618  * VdW interaction:            LennardJones
 619  * Geometry:                   Water4-Particle
 620  * Calculate force/pot:        Force
 621  */
 622 void
 623 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 624                     (t_nblist * gmx_restrict                nlist,
 625                      rvec * gmx_restrict                    xx,
 626                      rvec * gmx_restrict                    ff,
 627                      t_forcerec * gmx_restrict              fr,
 628                      t_mdatoms * gmx_restrict               mdatoms,
 629                      nb_kernel_data_t * gmx_restrict        kernel_data,
 630                      t_nrnb * gmx_restrict                  nrnb)
 631 {
 632     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 633      * just 0 for non-waters.
 634      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 635      * jnr indices corresponding to data put in the four positions in the SIMD register.
 636      */
 637     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 638     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 639     int              jnrA,jnrB,jnrC,jnrD;
 640     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 641     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 642     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 643     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 644     real             rcutoff_scalar;
 645     real             *shiftvec,*fshift,*x,*f;
 646     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 647     real             scratch[4*DIM];
 648     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 649     real *           vdwioffsetptr0;
 650     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 651     real *           vdwioffsetptr1;
 652     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 653     real *           vdwioffsetptr2;
 654     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 655     real *           vdwioffsetptr3;
 656     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 657     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 658     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 659     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 660     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 661     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 662     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 663     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 664     real             *charge;
 665     int              nvdwtype;
 666     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 667     int              *vdwtype;
 668     real             *vdwparam;
 669     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 670     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 671     __m256d          dummy_mask,cutoff_mask;
 672     __m128           tmpmask0,tmpmask1;
 673     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 674     __m256d          one     = _mm256_set1_pd(1.0);
 675     __m256d          two     = _mm256_set1_pd(2.0);
 676     x                = xx[0];
 677     f                = ff[0];
 678
 679     nri              = nlist->nri;
 680     iinr             = nlist->iinr;
 681     jindex           = nlist->jindex;
 682     jjnr             = nlist->jjnr;
 683     shiftidx         = nlist->shift;
 684     gid              = nlist->gid;
 685     shiftvec         = fr->shift_vec[0];
 686     fshift           = fr->fshift[0];
 687     facel            = _mm256_set1_pd(fr->epsfac);
 688     charge           = mdatoms->chargeA;
 689     nvdwtype         = fr->ntype;
 690     vdwparam         = fr->nbfp;
 691     vdwtype          = mdatoms->typeA;
 692
 693     /* Setup water-specific parameters */
 694     inr              = nlist->iinr[0];
 695     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 696     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 697     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 698     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 699
 700     /* Avoid stupid compiler warnings */
 701     jnrA = jnrB = jnrC = jnrD = 0;
 702     j_coord_offsetA = 0;
 703     j_coord_offsetB = 0;
 704     j_coord_offsetC = 0;
 705     j_coord_offsetD = 0;
 706
 707     outeriter        = 0;
 708     inneriter        = 0;
 709
 710     for(iidx=0;iidx<4*DIM;iidx++)
 711     {
 712         scratch[iidx] = 0.0;
 713     }
 714
 715     /* Start outer loop over neighborlists */
 716     for(iidx=0; iidx<nri; iidx++)
 717     {
 718         /* Load shift vector for this list */
 719         i_shift_offset   = DIM*shiftidx[iidx];
 720
 721         /* Load limits for loop over neighbors */
 722         j_index_start    = jindex[iidx];
 723         j_index_end      = jindex[iidx+1];
 724
 725         /* Get outer coordinate index */
 726         inr              = iinr[iidx];
 727         i_coord_offset   = DIM*inr;
 728
 729         /* Load i particle coords and add shift vector */
 730         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 731                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 732
 733         fix0             = _mm256_setzero_pd();
 734         fiy0             = _mm256_setzero_pd();
 735         fiz0             = _mm256_setzero_pd();
 736         fix1             = _mm256_setzero_pd();
 737         fiy1             = _mm256_setzero_pd();
 738         fiz1             = _mm256_setzero_pd();
 739         fix2             = _mm256_setzero_pd();
 740         fiy2             = _mm256_setzero_pd();
 741         fiz2             = _mm256_setzero_pd();
 742         fix3             = _mm256_setzero_pd();
 743         fiy3             = _mm256_setzero_pd();
 744         fiz3             = _mm256_setzero_pd();
 745
 746         /* Start inner kernel loop */
 747         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 748         {
 749
 750             /* Get j neighbor index, and coordinate index */
 751             jnrA             = jjnr[jidx];
 752             jnrB             = jjnr[jidx+1];
 753             jnrC             = jjnr[jidx+2];
 754             jnrD             = jjnr[jidx+3];
 755             j_coord_offsetA  = DIM*jnrA;
 756             j_coord_offsetB  = DIM*jnrB;
 757             j_coord_offsetC  = DIM*jnrC;
 758             j_coord_offsetD  = DIM*jnrD;
 759
 760             /* load j atom coordinates */
 761             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 762                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 763                                                  &jx0,&jy0,&jz0);
 764
 765             /* Calculate displacement vector */
 766             dx00             = _mm256_sub_pd(ix0,jx0);
 767             dy00             = _mm256_sub_pd(iy0,jy0);
 768             dz00             = _mm256_sub_pd(iz0,jz0);
 769             dx10             = _mm256_sub_pd(ix1,jx0);
 770             dy10             = _mm256_sub_pd(iy1,jy0);
 771             dz10             = _mm256_sub_pd(iz1,jz0);
 772             dx20             = _mm256_sub_pd(ix2,jx0);
 773             dy20             = _mm256_sub_pd(iy2,jy0);
 774             dz20             = _mm256_sub_pd(iz2,jz0);
 775             dx30             = _mm256_sub_pd(ix3,jx0);
 776             dy30             = _mm256_sub_pd(iy3,jy0);
 777             dz30             = _mm256_sub_pd(iz3,jz0);
 778
 779             /* Calculate squared distance and things based on it */
 780             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 781             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 782             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 783             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 784
 785             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 786             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 787             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 788
 789             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 790             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 791             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 792             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 793
 794             /* Load parameters for j particles */
 795             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 796                                                                  charge+jnrC+0,charge+jnrD+0);
 797             vdwjidx0A        = 2*vdwtype[jnrA+0];
 798             vdwjidx0B        = 2*vdwtype[jnrB+0];
 799             vdwjidx0C        = 2*vdwtype[jnrC+0];
 800             vdwjidx0D        = 2*vdwtype[jnrD+0];
 801
 802             fjx0             = _mm256_setzero_pd();
 803             fjy0             = _mm256_setzero_pd();
 804             fjz0             = _mm256_setzero_pd();
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             /* Compute parameters for interactions between i and j atoms */
 811             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 812                                             vdwioffsetptr0+vdwjidx0B,
 813                                             vdwioffsetptr0+vdwjidx0C,
 814                                             vdwioffsetptr0+vdwjidx0D,
 815                                             &c6_00,&c12_00);
 816
 817             /* LENNARD-JONES DISPERSION/REPULSION */
 818
 819             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 820             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 821
 822             fscal            = fvdw;
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm256_mul_pd(fscal,dx00);
 826             ty               = _mm256_mul_pd(fscal,dy00);
 827             tz               = _mm256_mul_pd(fscal,dz00);
 828
 829             /* Update vectorial force */
 830             fix0             = _mm256_add_pd(fix0,tx);
 831             fiy0             = _mm256_add_pd(fiy0,ty);
 832             fiz0             = _mm256_add_pd(fiz0,tz);
 833
 834             fjx0             = _mm256_add_pd(fjx0,tx);
 835             fjy0             = _mm256_add_pd(fjy0,ty);
 836             fjz0             = _mm256_add_pd(fjz0,tz);
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq10             = _mm256_mul_pd(iq1,jq0);
 844
 845             /* COULOMB ELECTROSTATICS */
 846             velec            = _mm256_mul_pd(qq10,rinv10);
 847             felec            = _mm256_mul_pd(velec,rinvsq10);
 848
 849             fscal            = felec;
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_pd(fscal,dx10);
 853             ty               = _mm256_mul_pd(fscal,dy10);
 854             tz               = _mm256_mul_pd(fscal,dz10);
 855
 856             /* Update vectorial force */
 857             fix1             = _mm256_add_pd(fix1,tx);
 858             fiy1             = _mm256_add_pd(fiy1,ty);
 859             fiz1             = _mm256_add_pd(fiz1,tz);
 860
 861             fjx0             = _mm256_add_pd(fjx0,tx);
 862             fjy0             = _mm256_add_pd(fjy0,ty);
 863             fjz0             = _mm256_add_pd(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq20             = _mm256_mul_pd(iq2,jq0);
 871
 872             /* COULOMB ELECTROSTATICS */
 873             velec            = _mm256_mul_pd(qq20,rinv20);
 874             felec            = _mm256_mul_pd(velec,rinvsq20);
 875
 876             fscal            = felec;
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm256_mul_pd(fscal,dx20);
 880             ty               = _mm256_mul_pd(fscal,dy20);
 881             tz               = _mm256_mul_pd(fscal,dz20);
 882
 883             /* Update vectorial force */
 884             fix2             = _mm256_add_pd(fix2,tx);
 885             fiy2             = _mm256_add_pd(fiy2,ty);
 886             fiz2             = _mm256_add_pd(fiz2,tz);
 887
 888             fjx0             = _mm256_add_pd(fjx0,tx);
 889             fjy0             = _mm256_add_pd(fjy0,ty);
 890             fjz0             = _mm256_add_pd(fjz0,tz);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq30             = _mm256_mul_pd(iq3,jq0);
 898
 899             /* COULOMB ELECTROSTATICS */
 900             velec            = _mm256_mul_pd(qq30,rinv30);
 901             felec            = _mm256_mul_pd(velec,rinvsq30);
 902
 903             fscal            = felec;
 904
 905             /* Calculate temporary vectorial force */
 906             tx               = _mm256_mul_pd(fscal,dx30);
 907             ty               = _mm256_mul_pd(fscal,dy30);
 908             tz               = _mm256_mul_pd(fscal,dz30);
 909
 910             /* Update vectorial force */
 911             fix3             = _mm256_add_pd(fix3,tx);
 912             fiy3             = _mm256_add_pd(fiy3,ty);
 913             fiz3             = _mm256_add_pd(fiz3,tz);
 914
 915             fjx0             = _mm256_add_pd(fjx0,tx);
 916             fjy0             = _mm256_add_pd(fjy0,ty);
 917             fjz0             = _mm256_add_pd(fjz0,tz);
 918
 919             fjptrA             = f+j_coord_offsetA;
 920             fjptrB             = f+j_coord_offsetB;
 921             fjptrC             = f+j_coord_offsetC;
 922             fjptrD             = f+j_coord_offsetD;
 923
 924             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 925
 926             /* Inner loop uses 108 flops */
 927         }
 928
 929         if(jidx<j_index_end)
 930         {
 931
 932             /* Get j neighbor index, and coordinate index */
 933             jnrlistA         = jjnr[jidx];
 934             jnrlistB         = jjnr[jidx+1];
 935             jnrlistC         = jjnr[jidx+2];
 936             jnrlistD         = jjnr[jidx+3];
 937             /* Sign of each element will be negative for non-real atoms.
 938              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 939              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 940              */
 941             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 942
 943             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 944             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 945             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 946
 947             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 948             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 949             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 950             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 951             j_coord_offsetA  = DIM*jnrA;
 952             j_coord_offsetB  = DIM*jnrB;
 953             j_coord_offsetC  = DIM*jnrC;
 954             j_coord_offsetD  = DIM*jnrD;
 955
 956             /* load j atom coordinates */
 957             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 958                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 959                                                  &jx0,&jy0,&jz0);
 960
 961             /* Calculate displacement vector */
 962             dx00             = _mm256_sub_pd(ix0,jx0);
 963             dy00             = _mm256_sub_pd(iy0,jy0);
 964             dz00             = _mm256_sub_pd(iz0,jz0);
 965             dx10             = _mm256_sub_pd(ix1,jx0);
 966             dy10             = _mm256_sub_pd(iy1,jy0);
 967             dz10             = _mm256_sub_pd(iz1,jz0);
 968             dx20             = _mm256_sub_pd(ix2,jx0);
 969             dy20             = _mm256_sub_pd(iy2,jy0);
 970             dz20             = _mm256_sub_pd(iz2,jz0);
 971             dx30             = _mm256_sub_pd(ix3,jx0);
 972             dy30             = _mm256_sub_pd(iy3,jy0);
 973             dz30             = _mm256_sub_pd(iz3,jz0);
 974
 975             /* Calculate squared distance and things based on it */
 976             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 977             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 978             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 979             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 980
 981             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 982             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 983             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 984
 985             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 986             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 987             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 988             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 989
 990             /* Load parameters for j particles */
 991             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 992                                                                  charge+jnrC+0,charge+jnrD+0);
 993             vdwjidx0A        = 2*vdwtype[jnrA+0];
 994             vdwjidx0B        = 2*vdwtype[jnrB+0];
 995             vdwjidx0C        = 2*vdwtype[jnrC+0];
 996             vdwjidx0D        = 2*vdwtype[jnrD+0];
 997
 998             fjx0             = _mm256_setzero_pd();
 999             fjy0             = _mm256_setzero_pd();
1000             fjz0             = _mm256_setzero_pd();
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             &c6_00,&c12_00);
1012
1013             /* LENNARD-JONES DISPERSION/REPULSION */
1014
1015             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1016             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1017
1018             fscal            = fvdw;
1019
1020             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm256_mul_pd(fscal,dx00);
1024             ty               = _mm256_mul_pd(fscal,dy00);
1025             tz               = _mm256_mul_pd(fscal,dz00);
1026
1027             /* Update vectorial force */
1028             fix0             = _mm256_add_pd(fix0,tx);
1029             fiy0             = _mm256_add_pd(fiy0,ty);
1030             fiz0             = _mm256_add_pd(fiz0,tz);
1031
1032             fjx0             = _mm256_add_pd(fjx0,tx);
1033             fjy0             = _mm256_add_pd(fjy0,ty);
1034             fjz0             = _mm256_add_pd(fjz0,tz);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq10             = _mm256_mul_pd(iq1,jq0);
1042
1043             /* COULOMB ELECTROSTATICS */
1044             velec            = _mm256_mul_pd(qq10,rinv10);
1045             felec            = _mm256_mul_pd(velec,rinvsq10);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm256_mul_pd(fscal,dx10);
1053             ty               = _mm256_mul_pd(fscal,dy10);
1054             tz               = _mm256_mul_pd(fscal,dz10);
1055
1056             /* Update vectorial force */
1057             fix1             = _mm256_add_pd(fix1,tx);
1058             fiy1             = _mm256_add_pd(fiy1,ty);
1059             fiz1             = _mm256_add_pd(fiz1,tz);
1060
1061             fjx0             = _mm256_add_pd(fjx0,tx);
1062             fjy0             = _mm256_add_pd(fjy0,ty);
1063             fjz0             = _mm256_add_pd(fjz0,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq20             = _mm256_mul_pd(iq2,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm256_mul_pd(qq20,rinv20);
1074             felec            = _mm256_mul_pd(velec,rinvsq20);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_pd(fscal,dx20);
1082             ty               = _mm256_mul_pd(fscal,dy20);
1083             tz               = _mm256_mul_pd(fscal,dz20);
1084
1085             /* Update vectorial force */
1086             fix2             = _mm256_add_pd(fix2,tx);
1087             fiy2             = _mm256_add_pd(fiy2,ty);
1088             fiz2             = _mm256_add_pd(fiz2,tz);
1089
1090             fjx0             = _mm256_add_pd(fjx0,tx);
1091             fjy0             = _mm256_add_pd(fjy0,ty);
1092             fjz0             = _mm256_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq30             = _mm256_mul_pd(iq3,jq0);
1100
1101             /* COULOMB ELECTROSTATICS */
1102             velec            = _mm256_mul_pd(qq30,rinv30);
1103             felec            = _mm256_mul_pd(velec,rinvsq30);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm256_mul_pd(fscal,dx30);
1111             ty               = _mm256_mul_pd(fscal,dy30);
1112             tz               = _mm256_mul_pd(fscal,dz30);
1113
1114             /* Update vectorial force */
1115             fix3             = _mm256_add_pd(fix3,tx);
1116             fiy3             = _mm256_add_pd(fiy3,ty);
1117             fiz3             = _mm256_add_pd(fiz3,tz);
1118
1119             fjx0             = _mm256_add_pd(fjx0,tx);
1120             fjy0             = _mm256_add_pd(fjy0,ty);
1121             fjz0             = _mm256_add_pd(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 108 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1136                                                  f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 24 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1150 }