src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single 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_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          ewitab;
  92     __m128i          ewitab_lo,ewitab_hi;
  93     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 120     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 121     beta2            = _mm256_mul_ps(beta,beta);
 122     beta3            = _mm256_mul_ps(beta,beta2);
 123
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 139
 140     rswitch_scalar   = fr->rcoulomb_switch;
 141     rswitch          = _mm256_set1_ps(rswitch_scalar);
 142     /* Setup switch parameters */
 143     d_scalar         = rcutoff_scalar-rswitch_scalar;
 144     d                = _mm256_set1_ps(d_scalar);
 145     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 146     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 149     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158     j_coord_offsetE = 0;
 159     j_coord_offsetF = 0;
 160     j_coord_offsetG = 0;
 161     j_coord_offsetH = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 188
 189         fix0             = _mm256_setzero_ps();
 190         fiy0             = _mm256_setzero_ps();
 191         fiz0             = _mm256_setzero_ps();
 192         fix1             = _mm256_setzero_ps();
 193         fiy1             = _mm256_setzero_ps();
 194         fiz1             = _mm256_setzero_ps();
 195         fix2             = _mm256_setzero_ps();
 196         fiy2             = _mm256_setzero_ps();
 197         fiz2             = _mm256_setzero_ps();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm256_setzero_ps();
 201         vvdwsum          = _mm256_setzero_ps();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             jnrE             = jjnr[jidx+4];
 213             jnrF             = jjnr[jidx+5];
 214             jnrG             = jjnr[jidx+6];
 215             jnrH             = jjnr[jidx+7];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220             j_coord_offsetE  = DIM*jnrE;
 221             j_coord_offsetF  = DIM*jnrF;
 222             j_coord_offsetG  = DIM*jnrG;
 223             j_coord_offsetH  = DIM*jnrH;
 224
 225             /* load j atom coordinates */
 226             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 228                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 229                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 230                                                  &jx0,&jy0,&jz0);
 231
 232             /* Calculate displacement vector */
 233             dx00             = _mm256_sub_ps(ix0,jx0);
 234             dy00             = _mm256_sub_ps(iy0,jy0);
 235             dz00             = _mm256_sub_ps(iz0,jz0);
 236             dx10             = _mm256_sub_ps(ix1,jx0);
 237             dy10             = _mm256_sub_ps(iy1,jy0);
 238             dz10             = _mm256_sub_ps(iz1,jz0);
 239             dx20             = _mm256_sub_ps(ix2,jx0);
 240             dy20             = _mm256_sub_ps(iy2,jy0);
 241             dz20             = _mm256_sub_ps(iz2,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 247
 248             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 249             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 250             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 251
 252             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 253             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 254             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0,
 259                                                                  charge+jnrE+0,charge+jnrF+0,
 260                                                                  charge+jnrG+0,charge+jnrH+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265             vdwjidx0E        = 2*vdwtype[jnrE+0];
 266             vdwjidx0F        = 2*vdwtype[jnrF+0];
 267             vdwjidx0G        = 2*vdwtype[jnrG+0];
 268             vdwjidx0H        = 2*vdwtype[jnrH+0];
 269
 270             fjx0             = _mm256_setzero_ps();
 271             fjy0             = _mm256_setzero_ps();
 272             fjz0             = _mm256_setzero_ps();
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 279             {
 280
 281             r00              = _mm256_mul_ps(rsq00,rinv00);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq00             = _mm256_mul_ps(iq0,jq0);
 285             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             vdwioffsetptr0+vdwjidx0E,
 290                                             vdwioffsetptr0+vdwjidx0F,
 291                                             vdwioffsetptr0+vdwjidx0G,
 292                                             vdwioffsetptr0+vdwjidx0H,
 293                                             &c6_00,&c12_00);
 294
 295             /* EWALD ELECTROSTATICS */
 296
 297             /* Analytical PME correction */
 298             zeta2            = _mm256_mul_ps(beta2,rsq00);
 299             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 300             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 301             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 302             felec            = _mm256_mul_ps(qq00,felec);
 303             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 304             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 305             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 306             velec            = _mm256_mul_ps(qq00,velec);
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 312             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 313             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 314             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 315
 316             d                = _mm256_sub_ps(r00,rswitch);
 317             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 318             d2               = _mm256_mul_ps(d,d);
 319             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 320
 321             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 322
 323             /* Evaluate switch function */
 324             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 325             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 326             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 327             velec            = _mm256_mul_ps(velec,sw);
 328             vvdw             = _mm256_mul_ps(vvdw,sw);
 329             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _mm256_and_ps(velec,cutoff_mask);
 333             velecsum         = _mm256_add_ps(velecsum,velec);
 334             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 335             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 336
 337             fscal            = _mm256_add_ps(felec,fvdw);
 338
 339             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm256_mul_ps(fscal,dx00);
 343             ty               = _mm256_mul_ps(fscal,dy00);
 344             tz               = _mm256_mul_ps(fscal,dz00);
 345
 346             /* Update vectorial force */
 347             fix0             = _mm256_add_ps(fix0,tx);
 348             fiy0             = _mm256_add_ps(fiy0,ty);
 349             fiz0             = _mm256_add_ps(fiz0,tz);
 350
 351             fjx0             = _mm256_add_ps(fjx0,tx);
 352             fjy0             = _mm256_add_ps(fjy0,ty);
 353             fjz0             = _mm256_add_ps(fjz0,tz);
 354
 355             }
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 362             {
 363
 364             r10              = _mm256_mul_ps(rsq10,rinv10);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq10             = _mm256_mul_ps(iq1,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Analytical PME correction */
 372             zeta2            = _mm256_mul_ps(beta2,rsq10);
 373             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 374             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 375             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 376             felec            = _mm256_mul_ps(qq10,felec);
 377             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 378             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 379             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 380             velec            = _mm256_mul_ps(qq10,velec);
 381
 382             d                = _mm256_sub_ps(r10,rswitch);
 383             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 384             d2               = _mm256_mul_ps(d,d);
 385             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 386
 387             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 388
 389             /* Evaluate switch function */
 390             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 391             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 392             velec            = _mm256_mul_ps(velec,sw);
 393             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velec            = _mm256_and_ps(velec,cutoff_mask);
 397             velecsum         = _mm256_add_ps(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 402
 403             /* Calculate temporary vectorial force */
 404             tx               = _mm256_mul_ps(fscal,dx10);
 405             ty               = _mm256_mul_ps(fscal,dy10);
 406             tz               = _mm256_mul_ps(fscal,dz10);
 407
 408             /* Update vectorial force */
 409             fix1             = _mm256_add_ps(fix1,tx);
 410             fiy1             = _mm256_add_ps(fiy1,ty);
 411             fiz1             = _mm256_add_ps(fiz1,tz);
 412
 413             fjx0             = _mm256_add_ps(fjx0,tx);
 414             fjy0             = _mm256_add_ps(fjy0,ty);
 415             fjz0             = _mm256_add_ps(fjz0,tz);
 416
 417             }
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 424             {
 425
 426             r20              = _mm256_mul_ps(rsq20,rinv20);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq20             = _mm256_mul_ps(iq2,jq0);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Analytical PME correction */
 434             zeta2            = _mm256_mul_ps(beta2,rsq20);
 435             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 436             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 437             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 438             felec            = _mm256_mul_ps(qq20,felec);
 439             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 440             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 441             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 442             velec            = _mm256_mul_ps(qq20,velec);
 443
 444             d                = _mm256_sub_ps(r20,rswitch);
 445             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 446             d2               = _mm256_mul_ps(d,d);
 447             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 448
 449             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 450
 451             /* Evaluate switch function */
 452             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 453             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 454             velec            = _mm256_mul_ps(velec,sw);
 455             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm256_and_ps(velec,cutoff_mask);
 459             velecsum         = _mm256_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm256_mul_ps(fscal,dx20);
 467             ty               = _mm256_mul_ps(fscal,dy20);
 468             tz               = _mm256_mul_ps(fscal,dz20);
 469
 470             /* Update vectorial force */
 471             fix2             = _mm256_add_ps(fix2,tx);
 472             fiy2             = _mm256_add_ps(fiy2,ty);
 473             fiz2             = _mm256_add_ps(fiz2,tz);
 474
 475             fjx0             = _mm256_add_ps(fjx0,tx);
 476             fjy0             = _mm256_add_ps(fjy0,ty);
 477             fjz0             = _mm256_add_ps(fjz0,tz);
 478
 479             }
 480
 481             fjptrA             = f+j_coord_offsetA;
 482             fjptrB             = f+j_coord_offsetB;
 483             fjptrC             = f+j_coord_offsetC;
 484             fjptrD             = f+j_coord_offsetD;
 485             fjptrE             = f+j_coord_offsetE;
 486             fjptrF             = f+j_coord_offsetF;
 487             fjptrG             = f+j_coord_offsetG;
 488             fjptrH             = f+j_coord_offsetH;
 489
 490             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 491
 492             /* Inner loop uses 345 flops */
 493         }
 494
 495         if(jidx<j_index_end)
 496         {
 497
 498             /* Get j neighbor index, and coordinate index */
 499             jnrlistA         = jjnr[jidx];
 500             jnrlistB         = jjnr[jidx+1];
 501             jnrlistC         = jjnr[jidx+2];
 502             jnrlistD         = jjnr[jidx+3];
 503             jnrlistE         = jjnr[jidx+4];
 504             jnrlistF         = jjnr[jidx+5];
 505             jnrlistG         = jjnr[jidx+6];
 506             jnrlistH         = jjnr[jidx+7];
 507             /* Sign of each element will be negative for non-real atoms.
 508              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 509              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 510              */
 511             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 512                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 513
 514             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 515             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 516             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 517             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 518             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 519             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 520             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 521             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 522             j_coord_offsetA  = DIM*jnrA;
 523             j_coord_offsetB  = DIM*jnrB;
 524             j_coord_offsetC  = DIM*jnrC;
 525             j_coord_offsetD  = DIM*jnrD;
 526             j_coord_offsetE  = DIM*jnrE;
 527             j_coord_offsetF  = DIM*jnrF;
 528             j_coord_offsetG  = DIM*jnrG;
 529             j_coord_offsetH  = DIM*jnrH;
 530
 531             /* load j atom coordinates */
 532             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 533                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 534                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 535                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 536                                                  &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm256_sub_ps(ix0,jx0);
 540             dy00             = _mm256_sub_ps(iy0,jy0);
 541             dz00             = _mm256_sub_ps(iz0,jz0);
 542             dx10             = _mm256_sub_ps(ix1,jx0);
 543             dy10             = _mm256_sub_ps(iy1,jy0);
 544             dz10             = _mm256_sub_ps(iz1,jz0);
 545             dx20             = _mm256_sub_ps(ix2,jx0);
 546             dy20             = _mm256_sub_ps(iy2,jy0);
 547             dz20             = _mm256_sub_ps(iz2,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 551             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 552             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 553
 554             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 555             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 556             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 557
 558             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 559             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 560             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 561
 562             /* Load parameters for j particles */
 563             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 564                                                                  charge+jnrC+0,charge+jnrD+0,
 565                                                                  charge+jnrE+0,charge+jnrF+0,
 566                                                                  charge+jnrG+0,charge+jnrH+0);
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568             vdwjidx0B        = 2*vdwtype[jnrB+0];
 569             vdwjidx0C        = 2*vdwtype[jnrC+0];
 570             vdwjidx0D        = 2*vdwtype[jnrD+0];
 571             vdwjidx0E        = 2*vdwtype[jnrE+0];
 572             vdwjidx0F        = 2*vdwtype[jnrF+0];
 573             vdwjidx0G        = 2*vdwtype[jnrG+0];
 574             vdwjidx0H        = 2*vdwtype[jnrH+0];
 575
 576             fjx0             = _mm256_setzero_ps();
 577             fjy0             = _mm256_setzero_ps();
 578             fjz0             = _mm256_setzero_ps();
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 585             {
 586
 587             r00              = _mm256_mul_ps(rsq00,rinv00);
 588             r00              = _mm256_andnot_ps(dummy_mask,r00);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq00             = _mm256_mul_ps(iq0,jq0);
 592             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 593                                             vdwioffsetptr0+vdwjidx0B,
 594                                             vdwioffsetptr0+vdwjidx0C,
 595                                             vdwioffsetptr0+vdwjidx0D,
 596                                             vdwioffsetptr0+vdwjidx0E,
 597                                             vdwioffsetptr0+vdwjidx0F,
 598                                             vdwioffsetptr0+vdwjidx0G,
 599                                             vdwioffsetptr0+vdwjidx0H,
 600                                             &c6_00,&c12_00);
 601
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Analytical PME correction */
 605             zeta2            = _mm256_mul_ps(beta2,rsq00);
 606             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 607             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 608             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 609             felec            = _mm256_mul_ps(qq00,felec);
 610             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 611             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 612             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 613             velec            = _mm256_mul_ps(qq00,velec);
 614
 615             /* LENNARD-JONES DISPERSION/REPULSION */
 616
 617             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 618             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 619             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 620             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 621             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 622
 623             d                = _mm256_sub_ps(r00,rswitch);
 624             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 625             d2               = _mm256_mul_ps(d,d);
 626             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 627
 628             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 629
 630             /* Evaluate switch function */
 631             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 632             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 633             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 634             velec            = _mm256_mul_ps(velec,sw);
 635             vvdw             = _mm256_mul_ps(vvdw,sw);
 636             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velec            = _mm256_and_ps(velec,cutoff_mask);
 640             velec            = _mm256_andnot_ps(dummy_mask,velec);
 641             velecsum         = _mm256_add_ps(velecsum,velec);
 642             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 643             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 644             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 645
 646             fscal            = _mm256_add_ps(felec,fvdw);
 647
 648             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 649
 650             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm256_mul_ps(fscal,dx00);
 654             ty               = _mm256_mul_ps(fscal,dy00);
 655             tz               = _mm256_mul_ps(fscal,dz00);
 656
 657             /* Update vectorial force */
 658             fix0             = _mm256_add_ps(fix0,tx);
 659             fiy0             = _mm256_add_ps(fiy0,ty);
 660             fiz0             = _mm256_add_ps(fiz0,tz);
 661
 662             fjx0             = _mm256_add_ps(fjx0,tx);
 663             fjy0             = _mm256_add_ps(fjy0,ty);
 664             fjz0             = _mm256_add_ps(fjz0,tz);
 665
 666             }
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 673             {
 674
 675             r10              = _mm256_mul_ps(rsq10,rinv10);
 676             r10              = _mm256_andnot_ps(dummy_mask,r10);
 677
 678             /* Compute parameters for interactions between i and j atoms */
 679             qq10             = _mm256_mul_ps(iq1,jq0);
 680
 681             /* EWALD ELECTROSTATICS */
 682
 683             /* Analytical PME correction */
 684             zeta2            = _mm256_mul_ps(beta2,rsq10);
 685             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 686             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 687             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 688             felec            = _mm256_mul_ps(qq10,felec);
 689             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 690             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 691             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 692             velec            = _mm256_mul_ps(qq10,velec);
 693
 694             d                = _mm256_sub_ps(r10,rswitch);
 695             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 696             d2               = _mm256_mul_ps(d,d);
 697             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 698
 699             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 700
 701             /* Evaluate switch function */
 702             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 703             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 704             velec            = _mm256_mul_ps(velec,sw);
 705             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_and_ps(velec,cutoff_mask);
 709             velec            = _mm256_andnot_ps(dummy_mask,velec);
 710             velecsum         = _mm256_add_ps(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 715
 716             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm256_mul_ps(fscal,dx10);
 720             ty               = _mm256_mul_ps(fscal,dy10);
 721             tz               = _mm256_mul_ps(fscal,dz10);
 722
 723             /* Update vectorial force */
 724             fix1             = _mm256_add_ps(fix1,tx);
 725             fiy1             = _mm256_add_ps(fiy1,ty);
 726             fiz1             = _mm256_add_ps(fiz1,tz);
 727
 728             fjx0             = _mm256_add_ps(fjx0,tx);
 729             fjy0             = _mm256_add_ps(fjy0,ty);
 730             fjz0             = _mm256_add_ps(fjz0,tz);
 731
 732             }
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 739             {
 740
 741             r20              = _mm256_mul_ps(rsq20,rinv20);
 742             r20              = _mm256_andnot_ps(dummy_mask,r20);
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq20             = _mm256_mul_ps(iq2,jq0);
 746
 747             /* EWALD ELECTROSTATICS */
 748
 749             /* Analytical PME correction */
 750             zeta2            = _mm256_mul_ps(beta2,rsq20);
 751             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 752             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 753             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 754             felec            = _mm256_mul_ps(qq20,felec);
 755             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 756             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 757             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 758             velec            = _mm256_mul_ps(qq20,velec);
 759
 760             d                = _mm256_sub_ps(r20,rswitch);
 761             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 762             d2               = _mm256_mul_ps(d,d);
 763             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 764
 765             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 766
 767             /* Evaluate switch function */
 768             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 769             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 770             velec            = _mm256_mul_ps(velec,sw);
 771             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 772
 773             /* Update potential sum for this i atom from the interaction with this j atom. */
 774             velec            = _mm256_and_ps(velec,cutoff_mask);
 775             velec            = _mm256_andnot_ps(dummy_mask,velec);
 776             velecsum         = _mm256_add_ps(velecsum,velec);
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 781
 782             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm256_mul_ps(fscal,dx20);
 786             ty               = _mm256_mul_ps(fscal,dy20);
 787             tz               = _mm256_mul_ps(fscal,dz20);
 788
 789             /* Update vectorial force */
 790             fix2             = _mm256_add_ps(fix2,tx);
 791             fiy2             = _mm256_add_ps(fiy2,ty);
 792             fiz2             = _mm256_add_ps(fiz2,tz);
 793
 794             fjx0             = _mm256_add_ps(fjx0,tx);
 795             fjy0             = _mm256_add_ps(fjy0,ty);
 796             fjz0             = _mm256_add_ps(fjz0,tz);
 797
 798             }
 799
 800             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 801             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 802             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 803             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 804             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 805             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 806             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 807             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 808
 809             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 810
 811             /* Inner loop uses 348 flops */
 812         }
 813
 814         /* End of innermost loop */
 815
 816         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 817                                                  f+i_coord_offset,fshift+i_shift_offset);
 818
 819         ggid                        = gid[iidx];
 820         /* Update potential energies */
 821         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 822         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 823
 824         /* Increment number of inner iterations */
 825         inneriter                  += j_index_end - j_index_start;
 826
 827         /* Outer loop uses 20 flops */
 828     }
 829
 830     /* Increment number of outer iterations */
 831     outeriter        += nri;
 832
 833     /* Update outer/inner flops */
 834
 835     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*348);
 836 }
 837 /*
 838  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_single
 839  * Electrostatics interaction: Ewald
 840  * VdW interaction:            LennardJones
 841  * Geometry:                   Water3-Particle
 842  * Calculate force/pot:        Force
 843  */
 844 void
 845 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_single
 846                     (t_nblist * gmx_restrict                nlist,
 847                      rvec * gmx_restrict                    xx,
 848                      rvec * gmx_restrict                    ff,
 849                      t_forcerec * gmx_restrict              fr,
 850                      t_mdatoms * gmx_restrict               mdatoms,
 851                      nb_kernel_data_t * gmx_restrict        kernel_data,
 852                      t_nrnb * gmx_restrict                  nrnb)
 853 {
 854     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 855      * just 0 for non-waters.
 856      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 857      * jnr indices corresponding to data put in the four positions in the SIMD register.
 858      */
 859     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 860     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 861     int              jnrA,jnrB,jnrC,jnrD;
 862     int              jnrE,jnrF,jnrG,jnrH;
 863     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 864     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 865     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 866     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 867     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 868     real             rcutoff_scalar;
 869     real             *shiftvec,*fshift,*x,*f;
 870     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 871     real             scratch[4*DIM];
 872     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 873     real *           vdwioffsetptr0;
 874     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 875     real *           vdwioffsetptr1;
 876     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 877     real *           vdwioffsetptr2;
 878     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 879     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 880     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 881     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 882     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 883     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 884     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 885     real             *charge;
 886     int              nvdwtype;
 887     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 888     int              *vdwtype;
 889     real             *vdwparam;
 890     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 891     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 892     __m256i          ewitab;
 893     __m128i          ewitab_lo,ewitab_hi;
 894     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 895     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 896     real             *ewtab;
 897     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 898     real             rswitch_scalar,d_scalar;
 899     __m256           dummy_mask,cutoff_mask;
 900     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 901     __m256           one     = _mm256_set1_ps(1.0);
 902     __m256           two     = _mm256_set1_ps(2.0);
 903     x                = xx[0];
 904     f                = ff[0];
 905
 906     nri              = nlist->nri;
 907     iinr             = nlist->iinr;
 908     jindex           = nlist->jindex;
 909     jjnr             = nlist->jjnr;
 910     shiftidx         = nlist->shift;
 911     gid              = nlist->gid;
 912     shiftvec         = fr->shift_vec[0];
 913     fshift           = fr->fshift[0];
 914     facel            = _mm256_set1_ps(fr->epsfac);
 915     charge           = mdatoms->chargeA;
 916     nvdwtype         = fr->ntype;
 917     vdwparam         = fr->nbfp;
 918     vdwtype          = mdatoms->typeA;
 919
 920     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 921     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 922     beta2            = _mm256_mul_ps(beta,beta);
 923     beta3            = _mm256_mul_ps(beta,beta2);
 924
 925     ewtab            = fr->ic->tabq_coul_FDV0;
 926     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 927     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 928
 929     /* Setup water-specific parameters */
 930     inr              = nlist->iinr[0];
 931     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 932     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 933     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 934     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 935
 936     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 937     rcutoff_scalar   = fr->rcoulomb;
 938     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 939     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 940
 941     rswitch_scalar   = fr->rcoulomb_switch;
 942     rswitch          = _mm256_set1_ps(rswitch_scalar);
 943     /* Setup switch parameters */
 944     d_scalar         = rcutoff_scalar-rswitch_scalar;
 945     d                = _mm256_set1_ps(d_scalar);
 946     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 947     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 948     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 949     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 950     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 951     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 952
 953     /* Avoid stupid compiler warnings */
 954     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 955     j_coord_offsetA = 0;
 956     j_coord_offsetB = 0;
 957     j_coord_offsetC = 0;
 958     j_coord_offsetD = 0;
 959     j_coord_offsetE = 0;
 960     j_coord_offsetF = 0;
 961     j_coord_offsetG = 0;
 962     j_coord_offsetH = 0;
 963
 964     outeriter        = 0;
 965     inneriter        = 0;
 966
 967     for(iidx=0;iidx<4*DIM;iidx++)
 968     {
 969         scratch[iidx] = 0.0;
 970     }
 971
 972     /* Start outer loop over neighborlists */
 973     for(iidx=0; iidx<nri; iidx++)
 974     {
 975         /* Load shift vector for this list */
 976         i_shift_offset   = DIM*shiftidx[iidx];
 977
 978         /* Load limits for loop over neighbors */
 979         j_index_start    = jindex[iidx];
 980         j_index_end      = jindex[iidx+1];
 981
 982         /* Get outer coordinate index */
 983         inr              = iinr[iidx];
 984         i_coord_offset   = DIM*inr;
 985
 986         /* Load i particle coords and add shift vector */
 987         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 988                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 989
 990         fix0             = _mm256_setzero_ps();
 991         fiy0             = _mm256_setzero_ps();
 992         fiz0             = _mm256_setzero_ps();
 993         fix1             = _mm256_setzero_ps();
 994         fiy1             = _mm256_setzero_ps();
 995         fiz1             = _mm256_setzero_ps();
 996         fix2             = _mm256_setzero_ps();
 997         fiy2             = _mm256_setzero_ps();
 998         fiz2             = _mm256_setzero_ps();
 999
1000         /* Start inner kernel loop */
1001         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1002         {
1003
1004             /* Get j neighbor index, and coordinate index */
1005             jnrA             = jjnr[jidx];
1006             jnrB             = jjnr[jidx+1];
1007             jnrC             = jjnr[jidx+2];
1008             jnrD             = jjnr[jidx+3];
1009             jnrE             = jjnr[jidx+4];
1010             jnrF             = jjnr[jidx+5];
1011             jnrG             = jjnr[jidx+6];
1012             jnrH             = jjnr[jidx+7];
1013             j_coord_offsetA  = DIM*jnrA;
1014             j_coord_offsetB  = DIM*jnrB;
1015             j_coord_offsetC  = DIM*jnrC;
1016             j_coord_offsetD  = DIM*jnrD;
1017             j_coord_offsetE  = DIM*jnrE;
1018             j_coord_offsetF  = DIM*jnrF;
1019             j_coord_offsetG  = DIM*jnrG;
1020             j_coord_offsetH  = DIM*jnrH;
1021
1022             /* load j atom coordinates */
1023             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1025                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1026                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1027                                                  &jx0,&jy0,&jz0);
1028
1029             /* Calculate displacement vector */
1030             dx00             = _mm256_sub_ps(ix0,jx0);
1031             dy00             = _mm256_sub_ps(iy0,jy0);
1032             dz00             = _mm256_sub_ps(iz0,jz0);
1033             dx10             = _mm256_sub_ps(ix1,jx0);
1034             dy10             = _mm256_sub_ps(iy1,jy0);
1035             dz10             = _mm256_sub_ps(iz1,jz0);
1036             dx20             = _mm256_sub_ps(ix2,jx0);
1037             dy20             = _mm256_sub_ps(iy2,jy0);
1038             dz20             = _mm256_sub_ps(iz2,jz0);
1039
1040             /* Calculate squared distance and things based on it */
1041             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1042             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1043             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1044
1045             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1046             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1047             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1048
1049             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1050             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1051             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1052
1053             /* Load parameters for j particles */
1054             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1055                                                                  charge+jnrC+0,charge+jnrD+0,
1056                                                                  charge+jnrE+0,charge+jnrF+0,
1057                                                                  charge+jnrG+0,charge+jnrH+0);
1058             vdwjidx0A        = 2*vdwtype[jnrA+0];
1059             vdwjidx0B        = 2*vdwtype[jnrB+0];
1060             vdwjidx0C        = 2*vdwtype[jnrC+0];
1061             vdwjidx0D        = 2*vdwtype[jnrD+0];
1062             vdwjidx0E        = 2*vdwtype[jnrE+0];
1063             vdwjidx0F        = 2*vdwtype[jnrF+0];
1064             vdwjidx0G        = 2*vdwtype[jnrG+0];
1065             vdwjidx0H        = 2*vdwtype[jnrH+0];
1066
1067             fjx0             = _mm256_setzero_ps();
1068             fjy0             = _mm256_setzero_ps();
1069             fjz0             = _mm256_setzero_ps();
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1076             {
1077
1078             r00              = _mm256_mul_ps(rsq00,rinv00);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq00             = _mm256_mul_ps(iq0,jq0);
1082             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1083                                             vdwioffsetptr0+vdwjidx0B,
1084                                             vdwioffsetptr0+vdwjidx0C,
1085                                             vdwioffsetptr0+vdwjidx0D,
1086                                             vdwioffsetptr0+vdwjidx0E,
1087                                             vdwioffsetptr0+vdwjidx0F,
1088                                             vdwioffsetptr0+vdwjidx0G,
1089                                             vdwioffsetptr0+vdwjidx0H,
1090                                             &c6_00,&c12_00);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Analytical PME correction */
1095             zeta2            = _mm256_mul_ps(beta2,rsq00);
1096             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1097             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1098             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1099             felec            = _mm256_mul_ps(qq00,felec);
1100             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1101             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1102             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1103             velec            = _mm256_mul_ps(qq00,velec);
1104
1105             /* LENNARD-JONES DISPERSION/REPULSION */
1106
1107             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1108             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1109             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1110             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1111             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1112
1113             d                = _mm256_sub_ps(r00,rswitch);
1114             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1115             d2               = _mm256_mul_ps(d,d);
1116             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1117
1118             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1119
1120             /* Evaluate switch function */
1121             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1122             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1123             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1124             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1125
1126             fscal            = _mm256_add_ps(felec,fvdw);
1127
1128             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm256_mul_ps(fscal,dx00);
1132             ty               = _mm256_mul_ps(fscal,dy00);
1133             tz               = _mm256_mul_ps(fscal,dz00);
1134
1135             /* Update vectorial force */
1136             fix0             = _mm256_add_ps(fix0,tx);
1137             fiy0             = _mm256_add_ps(fiy0,ty);
1138             fiz0             = _mm256_add_ps(fiz0,tz);
1139
1140             fjx0             = _mm256_add_ps(fjx0,tx);
1141             fjy0             = _mm256_add_ps(fjy0,ty);
1142             fjz0             = _mm256_add_ps(fjz0,tz);
1143
1144             }
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1151             {
1152
1153             r10              = _mm256_mul_ps(rsq10,rinv10);
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             qq10             = _mm256_mul_ps(iq1,jq0);
1157
1158             /* EWALD ELECTROSTATICS */
1159
1160             /* Analytical PME correction */
1161             zeta2            = _mm256_mul_ps(beta2,rsq10);
1162             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1163             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1164             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1165             felec            = _mm256_mul_ps(qq10,felec);
1166             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1167             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1168             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1169             velec            = _mm256_mul_ps(qq10,velec);
1170
1171             d                = _mm256_sub_ps(r10,rswitch);
1172             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1173             d2               = _mm256_mul_ps(d,d);
1174             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1175
1176             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1177
1178             /* Evaluate switch function */
1179             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1180             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1181             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm256_mul_ps(fscal,dx10);
1189             ty               = _mm256_mul_ps(fscal,dy10);
1190             tz               = _mm256_mul_ps(fscal,dz10);
1191
1192             /* Update vectorial force */
1193             fix1             = _mm256_add_ps(fix1,tx);
1194             fiy1             = _mm256_add_ps(fiy1,ty);
1195             fiz1             = _mm256_add_ps(fiz1,tz);
1196
1197             fjx0             = _mm256_add_ps(fjx0,tx);
1198             fjy0             = _mm256_add_ps(fjy0,ty);
1199             fjz0             = _mm256_add_ps(fjz0,tz);
1200
1201             }
1202
1203             /**************************
1204              * CALCULATE INTERACTIONS *
1205              **************************/
1206
1207             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1208             {
1209
1210             r20              = _mm256_mul_ps(rsq20,rinv20);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq20             = _mm256_mul_ps(iq2,jq0);
1214
1215             /* EWALD ELECTROSTATICS */
1216
1217             /* Analytical PME correction */
1218             zeta2            = _mm256_mul_ps(beta2,rsq20);
1219             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1220             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1221             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1222             felec            = _mm256_mul_ps(qq20,felec);
1223             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1224             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1225             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1226             velec            = _mm256_mul_ps(qq20,velec);
1227
1228             d                = _mm256_sub_ps(r20,rswitch);
1229             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1230             d2               = _mm256_mul_ps(d,d);
1231             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1232
1233             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1234
1235             /* Evaluate switch function */
1236             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1237             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1238             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm256_mul_ps(fscal,dx20);
1246             ty               = _mm256_mul_ps(fscal,dy20);
1247             tz               = _mm256_mul_ps(fscal,dz20);
1248
1249             /* Update vectorial force */
1250             fix2             = _mm256_add_ps(fix2,tx);
1251             fiy2             = _mm256_add_ps(fiy2,ty);
1252             fiz2             = _mm256_add_ps(fiz2,tz);
1253
1254             fjx0             = _mm256_add_ps(fjx0,tx);
1255             fjy0             = _mm256_add_ps(fjy0,ty);
1256             fjz0             = _mm256_add_ps(fjz0,tz);
1257
1258             }
1259
1260             fjptrA             = f+j_coord_offsetA;
1261             fjptrB             = f+j_coord_offsetB;
1262             fjptrC             = f+j_coord_offsetC;
1263             fjptrD             = f+j_coord_offsetD;
1264             fjptrE             = f+j_coord_offsetE;
1265             fjptrF             = f+j_coord_offsetF;
1266             fjptrG             = f+j_coord_offsetG;
1267             fjptrH             = f+j_coord_offsetH;
1268
1269             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1270
1271             /* Inner loop uses 333 flops */
1272         }
1273
1274         if(jidx<j_index_end)
1275         {
1276
1277             /* Get j neighbor index, and coordinate index */
1278             jnrlistA         = jjnr[jidx];
1279             jnrlistB         = jjnr[jidx+1];
1280             jnrlistC         = jjnr[jidx+2];
1281             jnrlistD         = jjnr[jidx+3];
1282             jnrlistE         = jjnr[jidx+4];
1283             jnrlistF         = jjnr[jidx+5];
1284             jnrlistG         = jjnr[jidx+6];
1285             jnrlistH         = jjnr[jidx+7];
1286             /* Sign of each element will be negative for non-real atoms.
1287              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1288              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1289              */
1290             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1291                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1292
1293             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1294             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1295             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1296             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1297             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1298             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1299             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1300             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1301             j_coord_offsetA  = DIM*jnrA;
1302             j_coord_offsetB  = DIM*jnrB;
1303             j_coord_offsetC  = DIM*jnrC;
1304             j_coord_offsetD  = DIM*jnrD;
1305             j_coord_offsetE  = DIM*jnrE;
1306             j_coord_offsetF  = DIM*jnrF;
1307             j_coord_offsetG  = DIM*jnrG;
1308             j_coord_offsetH  = DIM*jnrH;
1309
1310             /* load j atom coordinates */
1311             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1312                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1313                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1314                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1315                                                  &jx0,&jy0,&jz0);
1316
1317             /* Calculate displacement vector */
1318             dx00             = _mm256_sub_ps(ix0,jx0);
1319             dy00             = _mm256_sub_ps(iy0,jy0);
1320             dz00             = _mm256_sub_ps(iz0,jz0);
1321             dx10             = _mm256_sub_ps(ix1,jx0);
1322             dy10             = _mm256_sub_ps(iy1,jy0);
1323             dz10             = _mm256_sub_ps(iz1,jz0);
1324             dx20             = _mm256_sub_ps(ix2,jx0);
1325             dy20             = _mm256_sub_ps(iy2,jy0);
1326             dz20             = _mm256_sub_ps(iz2,jz0);
1327
1328             /* Calculate squared distance and things based on it */
1329             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1330             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1331             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1332
1333             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1334             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1335             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1336
1337             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1338             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1339             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1340
1341             /* Load parameters for j particles */
1342             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1343                                                                  charge+jnrC+0,charge+jnrD+0,
1344                                                                  charge+jnrE+0,charge+jnrF+0,
1345                                                                  charge+jnrG+0,charge+jnrH+0);
1346             vdwjidx0A        = 2*vdwtype[jnrA+0];
1347             vdwjidx0B        = 2*vdwtype[jnrB+0];
1348             vdwjidx0C        = 2*vdwtype[jnrC+0];
1349             vdwjidx0D        = 2*vdwtype[jnrD+0];
1350             vdwjidx0E        = 2*vdwtype[jnrE+0];
1351             vdwjidx0F        = 2*vdwtype[jnrF+0];
1352             vdwjidx0G        = 2*vdwtype[jnrG+0];
1353             vdwjidx0H        = 2*vdwtype[jnrH+0];
1354
1355             fjx0             = _mm256_setzero_ps();
1356             fjy0             = _mm256_setzero_ps();
1357             fjz0             = _mm256_setzero_ps();
1358
1359             /**************************
1360              * CALCULATE INTERACTIONS *
1361              **************************/
1362
1363             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1364             {
1365
1366             r00              = _mm256_mul_ps(rsq00,rinv00);
1367             r00              = _mm256_andnot_ps(dummy_mask,r00);
1368
1369             /* Compute parameters for interactions between i and j atoms */
1370             qq00             = _mm256_mul_ps(iq0,jq0);
1371             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1372                                             vdwioffsetptr0+vdwjidx0B,
1373                                             vdwioffsetptr0+vdwjidx0C,
1374                                             vdwioffsetptr0+vdwjidx0D,
1375                                             vdwioffsetptr0+vdwjidx0E,
1376                                             vdwioffsetptr0+vdwjidx0F,
1377                                             vdwioffsetptr0+vdwjidx0G,
1378                                             vdwioffsetptr0+vdwjidx0H,
1379                                             &c6_00,&c12_00);
1380
1381             /* EWALD ELECTROSTATICS */
1382
1383             /* Analytical PME correction */
1384             zeta2            = _mm256_mul_ps(beta2,rsq00);
1385             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1386             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1387             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1388             felec            = _mm256_mul_ps(qq00,felec);
1389             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1390             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1391             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1392             velec            = _mm256_mul_ps(qq00,velec);
1393
1394             /* LENNARD-JONES DISPERSION/REPULSION */
1395
1396             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1397             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1398             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1399             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1400             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1401
1402             d                = _mm256_sub_ps(r00,rswitch);
1403             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1404             d2               = _mm256_mul_ps(d,d);
1405             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1406
1407             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1408
1409             /* Evaluate switch function */
1410             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1411             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1412             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1413             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1414
1415             fscal            = _mm256_add_ps(felec,fvdw);
1416
1417             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1418
1419             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1420
1421             /* Calculate temporary vectorial force */
1422             tx               = _mm256_mul_ps(fscal,dx00);
1423             ty               = _mm256_mul_ps(fscal,dy00);
1424             tz               = _mm256_mul_ps(fscal,dz00);
1425
1426             /* Update vectorial force */
1427             fix0             = _mm256_add_ps(fix0,tx);
1428             fiy0             = _mm256_add_ps(fiy0,ty);
1429             fiz0             = _mm256_add_ps(fiz0,tz);
1430
1431             fjx0             = _mm256_add_ps(fjx0,tx);
1432             fjy0             = _mm256_add_ps(fjy0,ty);
1433             fjz0             = _mm256_add_ps(fjz0,tz);
1434
1435             }
1436
1437             /**************************
1438              * CALCULATE INTERACTIONS *
1439              **************************/
1440
1441             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1442             {
1443
1444             r10              = _mm256_mul_ps(rsq10,rinv10);
1445             r10              = _mm256_andnot_ps(dummy_mask,r10);
1446
1447             /* Compute parameters for interactions between i and j atoms */
1448             qq10             = _mm256_mul_ps(iq1,jq0);
1449
1450             /* EWALD ELECTROSTATICS */
1451
1452             /* Analytical PME correction */
1453             zeta2            = _mm256_mul_ps(beta2,rsq10);
1454             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1455             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1456             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1457             felec            = _mm256_mul_ps(qq10,felec);
1458             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1459             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1460             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1461             velec            = _mm256_mul_ps(qq10,velec);
1462
1463             d                = _mm256_sub_ps(r10,rswitch);
1464             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1465             d2               = _mm256_mul_ps(d,d);
1466             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1467
1468             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1469
1470             /* Evaluate switch function */
1471             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1472             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1473             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1474
1475             fscal            = felec;
1476
1477             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1478
1479             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1480
1481             /* Calculate temporary vectorial force */
1482             tx               = _mm256_mul_ps(fscal,dx10);
1483             ty               = _mm256_mul_ps(fscal,dy10);
1484             tz               = _mm256_mul_ps(fscal,dz10);
1485
1486             /* Update vectorial force */
1487             fix1             = _mm256_add_ps(fix1,tx);
1488             fiy1             = _mm256_add_ps(fiy1,ty);
1489             fiz1             = _mm256_add_ps(fiz1,tz);
1490
1491             fjx0             = _mm256_add_ps(fjx0,tx);
1492             fjy0             = _mm256_add_ps(fjy0,ty);
1493             fjz0             = _mm256_add_ps(fjz0,tz);
1494
1495             }
1496
1497             /**************************
1498              * CALCULATE INTERACTIONS *
1499              **************************/
1500
1501             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1502             {
1503
1504             r20              = _mm256_mul_ps(rsq20,rinv20);
1505             r20              = _mm256_andnot_ps(dummy_mask,r20);
1506
1507             /* Compute parameters for interactions between i and j atoms */
1508             qq20             = _mm256_mul_ps(iq2,jq0);
1509
1510             /* EWALD ELECTROSTATICS */
1511
1512             /* Analytical PME correction */
1513             zeta2            = _mm256_mul_ps(beta2,rsq20);
1514             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1515             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1516             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1517             felec            = _mm256_mul_ps(qq20,felec);
1518             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1519             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1520             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1521             velec            = _mm256_mul_ps(qq20,velec);
1522
1523             d                = _mm256_sub_ps(r20,rswitch);
1524             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1525             d2               = _mm256_mul_ps(d,d);
1526             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1527
1528             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1529
1530             /* Evaluate switch function */
1531             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1532             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1533             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1534
1535             fscal            = felec;
1536
1537             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1538
1539             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1540
1541             /* Calculate temporary vectorial force */
1542             tx               = _mm256_mul_ps(fscal,dx20);
1543             ty               = _mm256_mul_ps(fscal,dy20);
1544             tz               = _mm256_mul_ps(fscal,dz20);
1545
1546             /* Update vectorial force */
1547             fix2             = _mm256_add_ps(fix2,tx);
1548             fiy2             = _mm256_add_ps(fiy2,ty);
1549             fiz2             = _mm256_add_ps(fiz2,tz);
1550
1551             fjx0             = _mm256_add_ps(fjx0,tx);
1552             fjy0             = _mm256_add_ps(fjy0,ty);
1553             fjz0             = _mm256_add_ps(fjz0,tz);
1554
1555             }
1556
1557             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1558             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1559             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1560             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1561             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1562             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1563             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1564             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1565
1566             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1567
1568             /* Inner loop uses 336 flops */
1569         }
1570
1571         /* End of innermost loop */
1572
1573         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1574                                                  f+i_coord_offset,fshift+i_shift_offset);
1575
1576         /* Increment number of inner iterations */
1577         inneriter                  += j_index_end - j_index_start;
1578
1579         /* Outer loop uses 18 flops */
1580     }
1581
1582     /* Increment number of outer iterations */
1583     outeriter        += nri;
1584
1585     /* Update outer/inner flops */
1586
1587     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*336);
1588 }