src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwNone_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_VdwNone_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwNone_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     __m256i          ewitab;
  86     __m128i          ewitab_lo,ewitab_hi;
  87     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  88     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  89     real             *ewtab;
  90     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     __m256           dummy_mask,cutoff_mask;
  93     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  94     __m256           one     = _mm256_set1_ps(1.0);
  95     __m256           two     = _mm256_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm256_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 111     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 112     beta2            = _mm256_mul_ps(beta,beta);
 113     beta3            = _mm256_mul_ps(beta,beta2);
 114
 115     ewtab            = fr->ic->tabq_coul_FDV0;
 116     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 117     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 122     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 123     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 124
 125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 128     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 129
 130     rswitch_scalar   = fr->rcoulomb_switch;
 131     rswitch          = _mm256_set1_ps(rswitch_scalar);
 132     /* Setup switch parameters */
 133     d_scalar         = rcutoff_scalar-rswitch_scalar;
 134     d                = _mm256_set1_ps(d_scalar);
 135     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 136     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 137     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 138     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 139     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148     j_coord_offsetE = 0;
 149     j_coord_offsetF = 0;
 150     j_coord_offsetG = 0;
 151     j_coord_offsetH = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 178
 179         fix0             = _mm256_setzero_ps();
 180         fiy0             = _mm256_setzero_ps();
 181         fiz0             = _mm256_setzero_ps();
 182         fix1             = _mm256_setzero_ps();
 183         fiy1             = _mm256_setzero_ps();
 184         fiz1             = _mm256_setzero_ps();
 185         fix2             = _mm256_setzero_ps();
 186         fiy2             = _mm256_setzero_ps();
 187         fiz2             = _mm256_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250
 251             fjx0             = _mm256_setzero_ps();
 252             fjy0             = _mm256_setzero_ps();
 253             fjz0             = _mm256_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 260             {
 261
 262             r00              = _mm256_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm256_mul_ps(iq0,jq0);
 266
 267             /* EWALD ELECTROSTATICS */
 268
 269             /* Analytical PME correction */
 270             zeta2            = _mm256_mul_ps(beta2,rsq00);
 271             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 272             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 273             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 274             felec            = _mm256_mul_ps(qq00,felec);
 275             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 276             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 277             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 278             velec            = _mm256_mul_ps(qq00,velec);
 279
 280             d                = _mm256_sub_ps(r00,rswitch);
 281             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 282             d2               = _mm256_mul_ps(d,d);
 283             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)))))));
 284
 285             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 286
 287             /* Evaluate switch function */
 288             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 289             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 290             velec            = _mm256_mul_ps(velec,sw);
 291             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm256_and_ps(velec,cutoff_mask);
 295             velecsum         = _mm256_add_ps(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_ps(fscal,dx00);
 303             ty               = _mm256_mul_ps(fscal,dy00);
 304             tz               = _mm256_mul_ps(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm256_add_ps(fix0,tx);
 308             fiy0             = _mm256_add_ps(fiy0,ty);
 309             fiz0             = _mm256_add_ps(fiz0,tz);
 310
 311             fjx0             = _mm256_add_ps(fjx0,tx);
 312             fjy0             = _mm256_add_ps(fjy0,ty);
 313             fjz0             = _mm256_add_ps(fjz0,tz);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 322             {
 323
 324             r10              = _mm256_mul_ps(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm256_mul_ps(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Analytical PME correction */
 332             zeta2            = _mm256_mul_ps(beta2,rsq10);
 333             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 334             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 335             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 336             felec            = _mm256_mul_ps(qq10,felec);
 337             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 338             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 339             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 340             velec            = _mm256_mul_ps(qq10,velec);
 341
 342             d                = _mm256_sub_ps(r10,rswitch);
 343             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 344             d2               = _mm256_mul_ps(d,d);
 345             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)))))));
 346
 347             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 348
 349             /* Evaluate switch function */
 350             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 351             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 352             velec            = _mm256_mul_ps(velec,sw);
 353             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _mm256_and_ps(velec,cutoff_mask);
 357             velecsum         = _mm256_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_ps(fscal,dx10);
 365             ty               = _mm256_mul_ps(fscal,dy10);
 366             tz               = _mm256_mul_ps(fscal,dz10);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm256_add_ps(fix1,tx);
 370             fiy1             = _mm256_add_ps(fiy1,ty);
 371             fiz1             = _mm256_add_ps(fiz1,tz);
 372
 373             fjx0             = _mm256_add_ps(fjx0,tx);
 374             fjy0             = _mm256_add_ps(fjy0,ty);
 375             fjz0             = _mm256_add_ps(fjz0,tz);
 376
 377             }
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 384             {
 385
 386             r20              = _mm256_mul_ps(rsq20,rinv20);
 387
 388             /* Compute parameters for interactions between i and j atoms */
 389             qq20             = _mm256_mul_ps(iq2,jq0);
 390
 391             /* EWALD ELECTROSTATICS */
 392
 393             /* Analytical PME correction */
 394             zeta2            = _mm256_mul_ps(beta2,rsq20);
 395             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 396             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 397             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 398             felec            = _mm256_mul_ps(qq20,felec);
 399             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 400             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 401             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 402             velec            = _mm256_mul_ps(qq20,velec);
 403
 404             d                = _mm256_sub_ps(r20,rswitch);
 405             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 406             d2               = _mm256_mul_ps(d,d);
 407             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)))))));
 408
 409             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 410
 411             /* Evaluate switch function */
 412             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 413             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 414             velec            = _mm256_mul_ps(velec,sw);
 415             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm256_and_ps(velec,cutoff_mask);
 419             velecsum         = _mm256_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_ps(fscal,dx20);
 427             ty               = _mm256_mul_ps(fscal,dy20);
 428             tz               = _mm256_mul_ps(fscal,dz20);
 429
 430             /* Update vectorial force */
 431             fix2             = _mm256_add_ps(fix2,tx);
 432             fiy2             = _mm256_add_ps(fiy2,ty);
 433             fiz2             = _mm256_add_ps(fiz2,tz);
 434
 435             fjx0             = _mm256_add_ps(fjx0,tx);
 436             fjy0             = _mm256_add_ps(fjy0,ty);
 437             fjz0             = _mm256_add_ps(fjz0,tz);
 438
 439             }
 440
 441             fjptrA             = f+j_coord_offsetA;
 442             fjptrB             = f+j_coord_offsetB;
 443             fjptrC             = f+j_coord_offsetC;
 444             fjptrD             = f+j_coord_offsetD;
 445             fjptrE             = f+j_coord_offsetE;
 446             fjptrF             = f+j_coord_offsetF;
 447             fjptrG             = f+j_coord_offsetG;
 448             fjptrH             = f+j_coord_offsetH;
 449
 450             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 451
 452             /* Inner loop uses 327 flops */
 453         }
 454
 455         if(jidx<j_index_end)
 456         {
 457
 458             /* Get j neighbor index, and coordinate index */
 459             jnrlistA         = jjnr[jidx];
 460             jnrlistB         = jjnr[jidx+1];
 461             jnrlistC         = jjnr[jidx+2];
 462             jnrlistD         = jjnr[jidx+3];
 463             jnrlistE         = jjnr[jidx+4];
 464             jnrlistF         = jjnr[jidx+5];
 465             jnrlistG         = jjnr[jidx+6];
 466             jnrlistH         = jjnr[jidx+7];
 467             /* Sign of each element will be negative for non-real atoms.
 468              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 469              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 470              */
 471             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 472                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 473
 474             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 475             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 476             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 477             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 478             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 479             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 480             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 481             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 482             j_coord_offsetA  = DIM*jnrA;
 483             j_coord_offsetB  = DIM*jnrB;
 484             j_coord_offsetC  = DIM*jnrC;
 485             j_coord_offsetD  = DIM*jnrD;
 486             j_coord_offsetE  = DIM*jnrE;
 487             j_coord_offsetF  = DIM*jnrF;
 488             j_coord_offsetG  = DIM*jnrG;
 489             j_coord_offsetH  = DIM*jnrH;
 490
 491             /* load j atom coordinates */
 492             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 493                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 494                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 495                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 496                                                  &jx0,&jy0,&jz0);
 497
 498             /* Calculate displacement vector */
 499             dx00             = _mm256_sub_ps(ix0,jx0);
 500             dy00             = _mm256_sub_ps(iy0,jy0);
 501             dz00             = _mm256_sub_ps(iz0,jz0);
 502             dx10             = _mm256_sub_ps(ix1,jx0);
 503             dy10             = _mm256_sub_ps(iy1,jy0);
 504             dz10             = _mm256_sub_ps(iz1,jz0);
 505             dx20             = _mm256_sub_ps(ix2,jx0);
 506             dy20             = _mm256_sub_ps(iy2,jy0);
 507             dz20             = _mm256_sub_ps(iz2,jz0);
 508
 509             /* Calculate squared distance and things based on it */
 510             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 511             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 512             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 513
 514             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 515             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 516             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 517
 518             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 519             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 520             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                                  charge+jnrC+0,charge+jnrD+0,
 525                                                                  charge+jnrE+0,charge+jnrF+0,
 526                                                                  charge+jnrG+0,charge+jnrH+0);
 527
 528             fjx0             = _mm256_setzero_ps();
 529             fjy0             = _mm256_setzero_ps();
 530             fjz0             = _mm256_setzero_ps();
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 537             {
 538
 539             r00              = _mm256_mul_ps(rsq00,rinv00);
 540             r00              = _mm256_andnot_ps(dummy_mask,r00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq00             = _mm256_mul_ps(iq0,jq0);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Analytical PME correction */
 548             zeta2            = _mm256_mul_ps(beta2,rsq00);
 549             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 550             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 551             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 552             felec            = _mm256_mul_ps(qq00,felec);
 553             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 554             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 555             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 556             velec            = _mm256_mul_ps(qq00,velec);
 557
 558             d                = _mm256_sub_ps(r00,rswitch);
 559             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 560             d2               = _mm256_mul_ps(d,d);
 561             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)))))));
 562
 563             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 564
 565             /* Evaluate switch function */
 566             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 567             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 568             velec            = _mm256_mul_ps(velec,sw);
 569             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm256_and_ps(velec,cutoff_mask);
 573             velec            = _mm256_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm256_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 579
 580             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_ps(fscal,dx00);
 584             ty               = _mm256_mul_ps(fscal,dy00);
 585             tz               = _mm256_mul_ps(fscal,dz00);
 586
 587             /* Update vectorial force */
 588             fix0             = _mm256_add_ps(fix0,tx);
 589             fiy0             = _mm256_add_ps(fiy0,ty);
 590             fiz0             = _mm256_add_ps(fiz0,tz);
 591
 592             fjx0             = _mm256_add_ps(fjx0,tx);
 593             fjy0             = _mm256_add_ps(fjy0,ty);
 594             fjz0             = _mm256_add_ps(fjz0,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 603             {
 604
 605             r10              = _mm256_mul_ps(rsq10,rinv10);
 606             r10              = _mm256_andnot_ps(dummy_mask,r10);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq10             = _mm256_mul_ps(iq1,jq0);
 610
 611             /* EWALD ELECTROSTATICS */
 612
 613             /* Analytical PME correction */
 614             zeta2            = _mm256_mul_ps(beta2,rsq10);
 615             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 616             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 617             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 618             felec            = _mm256_mul_ps(qq10,felec);
 619             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 620             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 621             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 622             velec            = _mm256_mul_ps(qq10,velec);
 623
 624             d                = _mm256_sub_ps(r10,rswitch);
 625             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 626             d2               = _mm256_mul_ps(d,d);
 627             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)))))));
 628
 629             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 630
 631             /* Evaluate switch function */
 632             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 633             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 634             velec            = _mm256_mul_ps(velec,sw);
 635             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm256_and_ps(velec,cutoff_mask);
 639             velec            = _mm256_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm256_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 645
 646             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_ps(fscal,dx10);
 650             ty               = _mm256_mul_ps(fscal,dy10);
 651             tz               = _mm256_mul_ps(fscal,dz10);
 652
 653             /* Update vectorial force */
 654             fix1             = _mm256_add_ps(fix1,tx);
 655             fiy1             = _mm256_add_ps(fiy1,ty);
 656             fiz1             = _mm256_add_ps(fiz1,tz);
 657
 658             fjx0             = _mm256_add_ps(fjx0,tx);
 659             fjy0             = _mm256_add_ps(fjy0,ty);
 660             fjz0             = _mm256_add_ps(fjz0,tz);
 661
 662             }
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 669             {
 670
 671             r20              = _mm256_mul_ps(rsq20,rinv20);
 672             r20              = _mm256_andnot_ps(dummy_mask,r20);
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq20             = _mm256_mul_ps(iq2,jq0);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Analytical PME correction */
 680             zeta2            = _mm256_mul_ps(beta2,rsq20);
 681             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 682             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 683             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 684             felec            = _mm256_mul_ps(qq20,felec);
 685             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 686             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 687             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 688             velec            = _mm256_mul_ps(qq20,velec);
 689
 690             d                = _mm256_sub_ps(r20,rswitch);
 691             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 692             d2               = _mm256_mul_ps(d,d);
 693             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)))))));
 694
 695             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 696
 697             /* Evaluate switch function */
 698             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 699             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 700             velec            = _mm256_mul_ps(velec,sw);
 701             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velec            = _mm256_and_ps(velec,cutoff_mask);
 705             velec            = _mm256_andnot_ps(dummy_mask,velec);
 706             velecsum         = _mm256_add_ps(velecsum,velec);
 707
 708             fscal            = felec;
 709
 710             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 711
 712             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm256_mul_ps(fscal,dx20);
 716             ty               = _mm256_mul_ps(fscal,dy20);
 717             tz               = _mm256_mul_ps(fscal,dz20);
 718
 719             /* Update vectorial force */
 720             fix2             = _mm256_add_ps(fix2,tx);
 721             fiy2             = _mm256_add_ps(fiy2,ty);
 722             fiz2             = _mm256_add_ps(fiz2,tz);
 723
 724             fjx0             = _mm256_add_ps(fjx0,tx);
 725             fjy0             = _mm256_add_ps(fjy0,ty);
 726             fjz0             = _mm256_add_ps(fjz0,tz);
 727
 728             }
 729
 730             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 731             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 732             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 733             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 734             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 735             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 736             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 737             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 738
 739             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 740
 741             /* Inner loop uses 330 flops */
 742         }
 743
 744         /* End of innermost loop */
 745
 746         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 747                                                  f+i_coord_offset,fshift+i_shift_offset);
 748
 749         ggid                        = gid[iidx];
 750         /* Update potential energies */
 751         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 752
 753         /* Increment number of inner iterations */
 754         inneriter                  += j_index_end - j_index_start;
 755
 756         /* Outer loop uses 19 flops */
 757     }
 758
 759     /* Increment number of outer iterations */
 760     outeriter        += nri;
 761
 762     /* Update outer/inner flops */
 763
 764     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*330);
 765 }
 766 /*
 767  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 768  * Electrostatics interaction: Ewald
 769  * VdW interaction:            None
 770  * Geometry:                   Water3-Particle
 771  * Calculate force/pot:        Force
 772  */
 773 void
 774 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 775                     (t_nblist * gmx_restrict                nlist,
 776                      rvec * gmx_restrict                    xx,
 777                      rvec * gmx_restrict                    ff,
 778                      t_forcerec * gmx_restrict              fr,
 779                      t_mdatoms * gmx_restrict               mdatoms,
 780                      nb_kernel_data_t * gmx_restrict        kernel_data,
 781                      t_nrnb * gmx_restrict                  nrnb)
 782 {
 783     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 784      * just 0 for non-waters.
 785      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 786      * jnr indices corresponding to data put in the four positions in the SIMD register.
 787      */
 788     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 789     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 790     int              jnrA,jnrB,jnrC,jnrD;
 791     int              jnrE,jnrF,jnrG,jnrH;
 792     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 793     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 794     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 795     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 796     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 797     real             rcutoff_scalar;
 798     real             *shiftvec,*fshift,*x,*f;
 799     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 800     real             scratch[4*DIM];
 801     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 802     real *           vdwioffsetptr0;
 803     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 804     real *           vdwioffsetptr1;
 805     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 806     real *           vdwioffsetptr2;
 807     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 808     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 809     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 810     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 811     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 812     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 813     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 814     real             *charge;
 815     __m256i          ewitab;
 816     __m128i          ewitab_lo,ewitab_hi;
 817     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 818     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 819     real             *ewtab;
 820     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 821     real             rswitch_scalar,d_scalar;
 822     __m256           dummy_mask,cutoff_mask;
 823     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 824     __m256           one     = _mm256_set1_ps(1.0);
 825     __m256           two     = _mm256_set1_ps(2.0);
 826     x                = xx[0];
 827     f                = ff[0];
 828
 829     nri              = nlist->nri;
 830     iinr             = nlist->iinr;
 831     jindex           = nlist->jindex;
 832     jjnr             = nlist->jjnr;
 833     shiftidx         = nlist->shift;
 834     gid              = nlist->gid;
 835     shiftvec         = fr->shift_vec[0];
 836     fshift           = fr->fshift[0];
 837     facel            = _mm256_set1_ps(fr->epsfac);
 838     charge           = mdatoms->chargeA;
 839
 840     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 841     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 842     beta2            = _mm256_mul_ps(beta,beta);
 843     beta3            = _mm256_mul_ps(beta,beta2);
 844
 845     ewtab            = fr->ic->tabq_coul_FDV0;
 846     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 847     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 848
 849     /* Setup water-specific parameters */
 850     inr              = nlist->iinr[0];
 851     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 852     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 853     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 854
 855     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 856     rcutoff_scalar   = fr->rcoulomb;
 857     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 858     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 859
 860     rswitch_scalar   = fr->rcoulomb_switch;
 861     rswitch          = _mm256_set1_ps(rswitch_scalar);
 862     /* Setup switch parameters */
 863     d_scalar         = rcutoff_scalar-rswitch_scalar;
 864     d                = _mm256_set1_ps(d_scalar);
 865     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 866     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 867     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 868     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 869     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 870     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 871
 872     /* Avoid stupid compiler warnings */
 873     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 874     j_coord_offsetA = 0;
 875     j_coord_offsetB = 0;
 876     j_coord_offsetC = 0;
 877     j_coord_offsetD = 0;
 878     j_coord_offsetE = 0;
 879     j_coord_offsetF = 0;
 880     j_coord_offsetG = 0;
 881     j_coord_offsetH = 0;
 882
 883     outeriter        = 0;
 884     inneriter        = 0;
 885
 886     for(iidx=0;iidx<4*DIM;iidx++)
 887     {
 888         scratch[iidx] = 0.0;
 889     }
 890
 891     /* Start outer loop over neighborlists */
 892     for(iidx=0; iidx<nri; iidx++)
 893     {
 894         /* Load shift vector for this list */
 895         i_shift_offset   = DIM*shiftidx[iidx];
 896
 897         /* Load limits for loop over neighbors */
 898         j_index_start    = jindex[iidx];
 899         j_index_end      = jindex[iidx+1];
 900
 901         /* Get outer coordinate index */
 902         inr              = iinr[iidx];
 903         i_coord_offset   = DIM*inr;
 904
 905         /* Load i particle coords and add shift vector */
 906         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 907                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 908
 909         fix0             = _mm256_setzero_ps();
 910         fiy0             = _mm256_setzero_ps();
 911         fiz0             = _mm256_setzero_ps();
 912         fix1             = _mm256_setzero_ps();
 913         fiy1             = _mm256_setzero_ps();
 914         fiz1             = _mm256_setzero_ps();
 915         fix2             = _mm256_setzero_ps();
 916         fiy2             = _mm256_setzero_ps();
 917         fiz2             = _mm256_setzero_ps();
 918
 919         /* Start inner kernel loop */
 920         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 921         {
 922
 923             /* Get j neighbor index, and coordinate index */
 924             jnrA             = jjnr[jidx];
 925             jnrB             = jjnr[jidx+1];
 926             jnrC             = jjnr[jidx+2];
 927             jnrD             = jjnr[jidx+3];
 928             jnrE             = jjnr[jidx+4];
 929             jnrF             = jjnr[jidx+5];
 930             jnrG             = jjnr[jidx+6];
 931             jnrH             = jjnr[jidx+7];
 932             j_coord_offsetA  = DIM*jnrA;
 933             j_coord_offsetB  = DIM*jnrB;
 934             j_coord_offsetC  = DIM*jnrC;
 935             j_coord_offsetD  = DIM*jnrD;
 936             j_coord_offsetE  = DIM*jnrE;
 937             j_coord_offsetF  = DIM*jnrF;
 938             j_coord_offsetG  = DIM*jnrG;
 939             j_coord_offsetH  = DIM*jnrH;
 940
 941             /* load j atom coordinates */
 942             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 943                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 944                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 945                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 946                                                  &jx0,&jy0,&jz0);
 947
 948             /* Calculate displacement vector */
 949             dx00             = _mm256_sub_ps(ix0,jx0);
 950             dy00             = _mm256_sub_ps(iy0,jy0);
 951             dz00             = _mm256_sub_ps(iz0,jz0);
 952             dx10             = _mm256_sub_ps(ix1,jx0);
 953             dy10             = _mm256_sub_ps(iy1,jy0);
 954             dz10             = _mm256_sub_ps(iz1,jz0);
 955             dx20             = _mm256_sub_ps(ix2,jx0);
 956             dy20             = _mm256_sub_ps(iy2,jy0);
 957             dz20             = _mm256_sub_ps(iz2,jz0);
 958
 959             /* Calculate squared distance and things based on it */
 960             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 961             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 962             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 963
 964             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 965             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 966             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 967
 968             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 969             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 970             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 971
 972             /* Load parameters for j particles */
 973             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 974                                                                  charge+jnrC+0,charge+jnrD+0,
 975                                                                  charge+jnrE+0,charge+jnrF+0,
 976                                                                  charge+jnrG+0,charge+jnrH+0);
 977
 978             fjx0             = _mm256_setzero_ps();
 979             fjy0             = _mm256_setzero_ps();
 980             fjz0             = _mm256_setzero_ps();
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 987             {
 988
 989             r00              = _mm256_mul_ps(rsq00,rinv00);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq00             = _mm256_mul_ps(iq0,jq0);
 993
 994             /* EWALD ELECTROSTATICS */
 995
 996             /* Analytical PME correction */
 997             zeta2            = _mm256_mul_ps(beta2,rsq00);
 998             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 999             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1000             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1001             felec            = _mm256_mul_ps(qq00,felec);
1002             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1003             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1004             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1005             velec            = _mm256_mul_ps(qq00,velec);
1006
1007             d                = _mm256_sub_ps(r00,rswitch);
1008             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1009             d2               = _mm256_mul_ps(d,d);
1010             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)))))));
1011
1012             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1013
1014             /* Evaluate switch function */
1015             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1016             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1017             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm256_mul_ps(fscal,dx00);
1025             ty               = _mm256_mul_ps(fscal,dy00);
1026             tz               = _mm256_mul_ps(fscal,dz00);
1027
1028             /* Update vectorial force */
1029             fix0             = _mm256_add_ps(fix0,tx);
1030             fiy0             = _mm256_add_ps(fiy0,ty);
1031             fiz0             = _mm256_add_ps(fiz0,tz);
1032
1033             fjx0             = _mm256_add_ps(fjx0,tx);
1034             fjy0             = _mm256_add_ps(fjy0,ty);
1035             fjz0             = _mm256_add_ps(fjz0,tz);
1036
1037             }
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1044             {
1045
1046             r10              = _mm256_mul_ps(rsq10,rinv10);
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq10             = _mm256_mul_ps(iq1,jq0);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Analytical PME correction */
1054             zeta2            = _mm256_mul_ps(beta2,rsq10);
1055             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1056             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1057             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1058             felec            = _mm256_mul_ps(qq10,felec);
1059             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1060             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1061             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1062             velec            = _mm256_mul_ps(qq10,velec);
1063
1064             d                = _mm256_sub_ps(r10,rswitch);
1065             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1066             d2               = _mm256_mul_ps(d,d);
1067             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)))))));
1068
1069             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1070
1071             /* Evaluate switch function */
1072             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1073             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1074             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_ps(fscal,dx10);
1082             ty               = _mm256_mul_ps(fscal,dy10);
1083             tz               = _mm256_mul_ps(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm256_add_ps(fix1,tx);
1087             fiy1             = _mm256_add_ps(fiy1,ty);
1088             fiz1             = _mm256_add_ps(fiz1,tz);
1089
1090             fjx0             = _mm256_add_ps(fjx0,tx);
1091             fjy0             = _mm256_add_ps(fjy0,ty);
1092             fjz0             = _mm256_add_ps(fjz0,tz);
1093
1094             }
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1101             {
1102
1103             r20              = _mm256_mul_ps(rsq20,rinv20);
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             qq20             = _mm256_mul_ps(iq2,jq0);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm256_mul_ps(beta2,rsq20);
1112             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1113             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1114             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1115             felec            = _mm256_mul_ps(qq20,felec);
1116             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1117             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1118             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1119             velec            = _mm256_mul_ps(qq20,velec);
1120
1121             d                = _mm256_sub_ps(r20,rswitch);
1122             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1123             d2               = _mm256_mul_ps(d,d);
1124             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)))))));
1125
1126             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1127
1128             /* Evaluate switch function */
1129             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1130             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1131             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm256_mul_ps(fscal,dx20);
1139             ty               = _mm256_mul_ps(fscal,dy20);
1140             tz               = _mm256_mul_ps(fscal,dz20);
1141
1142             /* Update vectorial force */
1143             fix2             = _mm256_add_ps(fix2,tx);
1144             fiy2             = _mm256_add_ps(fiy2,ty);
1145             fiz2             = _mm256_add_ps(fiz2,tz);
1146
1147             fjx0             = _mm256_add_ps(fjx0,tx);
1148             fjy0             = _mm256_add_ps(fjy0,ty);
1149             fjz0             = _mm256_add_ps(fjz0,tz);
1150
1151             }
1152
1153             fjptrA             = f+j_coord_offsetA;
1154             fjptrB             = f+j_coord_offsetB;
1155             fjptrC             = f+j_coord_offsetC;
1156             fjptrD             = f+j_coord_offsetD;
1157             fjptrE             = f+j_coord_offsetE;
1158             fjptrF             = f+j_coord_offsetF;
1159             fjptrG             = f+j_coord_offsetG;
1160             fjptrH             = f+j_coord_offsetH;
1161
1162             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1163
1164             /* Inner loop uses 318 flops */
1165         }
1166
1167         if(jidx<j_index_end)
1168         {
1169
1170             /* Get j neighbor index, and coordinate index */
1171             jnrlistA         = jjnr[jidx];
1172             jnrlistB         = jjnr[jidx+1];
1173             jnrlistC         = jjnr[jidx+2];
1174             jnrlistD         = jjnr[jidx+3];
1175             jnrlistE         = jjnr[jidx+4];
1176             jnrlistF         = jjnr[jidx+5];
1177             jnrlistG         = jjnr[jidx+6];
1178             jnrlistH         = jjnr[jidx+7];
1179             /* Sign of each element will be negative for non-real atoms.
1180              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1181              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1182              */
1183             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1184                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1185
1186             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1187             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1188             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1189             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1190             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1191             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1192             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1193             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1194             j_coord_offsetA  = DIM*jnrA;
1195             j_coord_offsetB  = DIM*jnrB;
1196             j_coord_offsetC  = DIM*jnrC;
1197             j_coord_offsetD  = DIM*jnrD;
1198             j_coord_offsetE  = DIM*jnrE;
1199             j_coord_offsetF  = DIM*jnrF;
1200             j_coord_offsetG  = DIM*jnrG;
1201             j_coord_offsetH  = DIM*jnrH;
1202
1203             /* load j atom coordinates */
1204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1208                                                  &jx0,&jy0,&jz0);
1209
1210             /* Calculate displacement vector */
1211             dx00             = _mm256_sub_ps(ix0,jx0);
1212             dy00             = _mm256_sub_ps(iy0,jy0);
1213             dz00             = _mm256_sub_ps(iz0,jz0);
1214             dx10             = _mm256_sub_ps(ix1,jx0);
1215             dy10             = _mm256_sub_ps(iy1,jy0);
1216             dz10             = _mm256_sub_ps(iz1,jz0);
1217             dx20             = _mm256_sub_ps(ix2,jx0);
1218             dy20             = _mm256_sub_ps(iy2,jy0);
1219             dz20             = _mm256_sub_ps(iz2,jz0);
1220
1221             /* Calculate squared distance and things based on it */
1222             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1223             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1224             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1225
1226             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1227             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1228             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1229
1230             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1231             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1232             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1233
1234             /* Load parameters for j particles */
1235             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1236                                                                  charge+jnrC+0,charge+jnrD+0,
1237                                                                  charge+jnrE+0,charge+jnrF+0,
1238                                                                  charge+jnrG+0,charge+jnrH+0);
1239
1240             fjx0             = _mm256_setzero_ps();
1241             fjy0             = _mm256_setzero_ps();
1242             fjz0             = _mm256_setzero_ps();
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1249             {
1250
1251             r00              = _mm256_mul_ps(rsq00,rinv00);
1252             r00              = _mm256_andnot_ps(dummy_mask,r00);
1253
1254             /* Compute parameters for interactions between i and j atoms */
1255             qq00             = _mm256_mul_ps(iq0,jq0);
1256
1257             /* EWALD ELECTROSTATICS */
1258
1259             /* Analytical PME correction */
1260             zeta2            = _mm256_mul_ps(beta2,rsq00);
1261             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1262             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1263             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1264             felec            = _mm256_mul_ps(qq00,felec);
1265             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1266             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1267             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1268             velec            = _mm256_mul_ps(qq00,velec);
1269
1270             d                = _mm256_sub_ps(r00,rswitch);
1271             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1272             d2               = _mm256_mul_ps(d,d);
1273             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)))))));
1274
1275             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1276
1277             /* Evaluate switch function */
1278             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1279             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1280             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1281
1282             fscal            = felec;
1283
1284             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1285
1286             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1287
1288             /* Calculate temporary vectorial force */
1289             tx               = _mm256_mul_ps(fscal,dx00);
1290             ty               = _mm256_mul_ps(fscal,dy00);
1291             tz               = _mm256_mul_ps(fscal,dz00);
1292
1293             /* Update vectorial force */
1294             fix0             = _mm256_add_ps(fix0,tx);
1295             fiy0             = _mm256_add_ps(fiy0,ty);
1296             fiz0             = _mm256_add_ps(fiz0,tz);
1297
1298             fjx0             = _mm256_add_ps(fjx0,tx);
1299             fjy0             = _mm256_add_ps(fjy0,ty);
1300             fjz0             = _mm256_add_ps(fjz0,tz);
1301
1302             }
1303
1304             /**************************
1305              * CALCULATE INTERACTIONS *
1306              **************************/
1307
1308             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1309             {
1310
1311             r10              = _mm256_mul_ps(rsq10,rinv10);
1312             r10              = _mm256_andnot_ps(dummy_mask,r10);
1313
1314             /* Compute parameters for interactions between i and j atoms */
1315             qq10             = _mm256_mul_ps(iq1,jq0);
1316
1317             /* EWALD ELECTROSTATICS */
1318
1319             /* Analytical PME correction */
1320             zeta2            = _mm256_mul_ps(beta2,rsq10);
1321             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1322             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1324             felec            = _mm256_mul_ps(qq10,felec);
1325             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1326             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1327             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1328             velec            = _mm256_mul_ps(qq10,velec);
1329
1330             d                = _mm256_sub_ps(r10,rswitch);
1331             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1332             d2               = _mm256_mul_ps(d,d);
1333             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)))))));
1334
1335             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1336
1337             /* Evaluate switch function */
1338             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1339             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1340             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1341
1342             fscal            = felec;
1343
1344             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1345
1346             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1347
1348             /* Calculate temporary vectorial force */
1349             tx               = _mm256_mul_ps(fscal,dx10);
1350             ty               = _mm256_mul_ps(fscal,dy10);
1351             tz               = _mm256_mul_ps(fscal,dz10);
1352
1353             /* Update vectorial force */
1354             fix1             = _mm256_add_ps(fix1,tx);
1355             fiy1             = _mm256_add_ps(fiy1,ty);
1356             fiz1             = _mm256_add_ps(fiz1,tz);
1357
1358             fjx0             = _mm256_add_ps(fjx0,tx);
1359             fjy0             = _mm256_add_ps(fjy0,ty);
1360             fjz0             = _mm256_add_ps(fjz0,tz);
1361
1362             }
1363
1364             /**************************
1365              * CALCULATE INTERACTIONS *
1366              **************************/
1367
1368             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1369             {
1370
1371             r20              = _mm256_mul_ps(rsq20,rinv20);
1372             r20              = _mm256_andnot_ps(dummy_mask,r20);
1373
1374             /* Compute parameters for interactions between i and j atoms */
1375             qq20             = _mm256_mul_ps(iq2,jq0);
1376
1377             /* EWALD ELECTROSTATICS */
1378
1379             /* Analytical PME correction */
1380             zeta2            = _mm256_mul_ps(beta2,rsq20);
1381             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1382             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1383             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1384             felec            = _mm256_mul_ps(qq20,felec);
1385             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1386             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1387             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1388             velec            = _mm256_mul_ps(qq20,velec);
1389
1390             d                = _mm256_sub_ps(r20,rswitch);
1391             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1392             d2               = _mm256_mul_ps(d,d);
1393             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)))))));
1394
1395             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1396
1397             /* Evaluate switch function */
1398             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1399             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1400             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1401
1402             fscal            = felec;
1403
1404             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1405
1406             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1407
1408             /* Calculate temporary vectorial force */
1409             tx               = _mm256_mul_ps(fscal,dx20);
1410             ty               = _mm256_mul_ps(fscal,dy20);
1411             tz               = _mm256_mul_ps(fscal,dz20);
1412
1413             /* Update vectorial force */
1414             fix2             = _mm256_add_ps(fix2,tx);
1415             fiy2             = _mm256_add_ps(fiy2,ty);
1416             fiz2             = _mm256_add_ps(fiz2,tz);
1417
1418             fjx0             = _mm256_add_ps(fjx0,tx);
1419             fjy0             = _mm256_add_ps(fjy0,ty);
1420             fjz0             = _mm256_add_ps(fjz0,tz);
1421
1422             }
1423
1424             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1425             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1426             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1427             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1428             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1429             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1430             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1431             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1432
1433             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1434
1435             /* Inner loop uses 321 flops */
1436         }
1437
1438         /* End of innermost loop */
1439
1440         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1441                                                  f+i_coord_offset,fshift+i_shift_offset);
1442
1443         /* Increment number of inner iterations */
1444         inneriter                  += j_index_end - j_index_start;
1445
1446         /* Outer loop uses 18 flops */
1447     }
1448
1449     /* Increment number of outer iterations */
1450     outeriter        += nri;
1451
1452     /* Update outer/inner flops */
1453
1454     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*321);
1455 }