src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m256i          ewitab;
  99     __m128i          ewitab_lo,ewitab_hi;
 100     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 102     real             *ewtab;
 103     __m256           dummy_mask,cutoff_mask;
 104     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 105     __m256           one     = _mm256_set1_ps(1.0);
 106     __m256           two     = _mm256_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm256_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120
 121     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 122     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 123     beta2            = _mm256_mul_ps(beta,beta);
 124     beta3            = _mm256_mul_ps(beta,beta2);
 125
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 133     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 134     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->ic->rcoulomb;
 138     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             jnrE             = jjnr[jidx+4];
 201             jnrF             = jjnr[jidx+5];
 202             jnrG             = jjnr[jidx+6];
 203             jnrH             = jjnr[jidx+7];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208             j_coord_offsetE  = DIM*jnrE;
 209             j_coord_offsetF  = DIM*jnrF;
 210             j_coord_offsetG  = DIM*jnrG;
 211             j_coord_offsetH  = DIM*jnrH;
 212
 213             /* load j atom coordinates */
 214             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 216                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 217                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_ps(ix0,jx0);
 222             dy00             = _mm256_sub_ps(iy0,jy0);
 223             dz00             = _mm256_sub_ps(iz0,jz0);
 224             dx10             = _mm256_sub_ps(ix1,jx0);
 225             dy10             = _mm256_sub_ps(iy1,jy0);
 226             dz10             = _mm256_sub_ps(iz1,jz0);
 227             dx20             = _mm256_sub_ps(ix2,jx0);
 228             dy20             = _mm256_sub_ps(iy2,jy0);
 229             dz20             = _mm256_sub_ps(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 235
 236             rinv00           = avx256_invsqrt_f(rsq00);
 237             rinv10           = avx256_invsqrt_f(rsq10);
 238             rinv20           = avx256_invsqrt_f(rsq20);
 239
 240             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 241             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 246                                                                  charge+jnrC+0,charge+jnrD+0,
 247                                                                  charge+jnrE+0,charge+jnrF+0,
 248                                                                  charge+jnrG+0,charge+jnrH+0);
 249
 250             fjx0             = _mm256_setzero_ps();
 251             fjy0             = _mm256_setzero_ps();
 252             fjz0             = _mm256_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             r00              = _mm256_mul_ps(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq00             = _mm256_mul_ps(iq0,jq0);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Analytical PME correction */
 269             zeta2            = _mm256_mul_ps(beta2,rsq00);
 270             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 271             pmecorrF         = avx256_pmecorrF_f(zeta2);
 272             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 273             felec            = _mm256_mul_ps(qq00,felec);
 274             pmecorrV         = avx256_pmecorrV_f(zeta2);
 275             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 276             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 277             velec            = _mm256_mul_ps(qq00,velec);
 278
 279             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm256_and_ps(velec,cutoff_mask);
 283             velecsum         = _mm256_add_ps(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_ps(fscal,dx00);
 291             ty               = _mm256_mul_ps(fscal,dy00);
 292             tz               = _mm256_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_ps(fix0,tx);
 296             fiy0             = _mm256_add_ps(fiy0,ty);
 297             fiz0             = _mm256_add_ps(fiz0,tz);
 298
 299             fjx0             = _mm256_add_ps(fjx0,tx);
 300             fjy0             = _mm256_add_ps(fjy0,ty);
 301             fjz0             = _mm256_add_ps(fjz0,tz);
 302
 303             }
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             r10              = _mm256_mul_ps(rsq10,rinv10);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm256_mul_ps(iq1,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Analytical PME correction */
 320             zeta2            = _mm256_mul_ps(beta2,rsq10);
 321             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 322             pmecorrF         = avx256_pmecorrF_f(zeta2);
 323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 324             felec            = _mm256_mul_ps(qq10,felec);
 325             pmecorrV         = avx256_pmecorrV_f(zeta2);
 326             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 327             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 328             velec            = _mm256_mul_ps(qq10,velec);
 329
 330             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm256_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm256_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_ps(fscal,dx10);
 342             ty               = _mm256_mul_ps(fscal,dy10);
 343             tz               = _mm256_mul_ps(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm256_add_ps(fix1,tx);
 347             fiy1             = _mm256_add_ps(fiy1,ty);
 348             fiz1             = _mm256_add_ps(fiz1,tz);
 349
 350             fjx0             = _mm256_add_ps(fjx0,tx);
 351             fjy0             = _mm256_add_ps(fjy0,ty);
 352             fjz0             = _mm256_add_ps(fjz0,tz);
 353
 354             }
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 361             {
 362
 363             r20              = _mm256_mul_ps(rsq20,rinv20);
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq20             = _mm256_mul_ps(iq2,jq0);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Analytical PME correction */
 371             zeta2            = _mm256_mul_ps(beta2,rsq20);
 372             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 373             pmecorrF         = avx256_pmecorrF_f(zeta2);
 374             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 375             felec            = _mm256_mul_ps(qq20,felec);
 376             pmecorrV         = avx256_pmecorrV_f(zeta2);
 377             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 378             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 379             velec            = _mm256_mul_ps(qq20,velec);
 380
 381             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm256_and_ps(velec,cutoff_mask);
 385             velecsum         = _mm256_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_ps(fscal,dx20);
 393             ty               = _mm256_mul_ps(fscal,dy20);
 394             tz               = _mm256_mul_ps(fscal,dz20);
 395
 396             /* Update vectorial force */
 397             fix2             = _mm256_add_ps(fix2,tx);
 398             fiy2             = _mm256_add_ps(fiy2,ty);
 399             fiz2             = _mm256_add_ps(fiz2,tz);
 400
 401             fjx0             = _mm256_add_ps(fjx0,tx);
 402             fjy0             = _mm256_add_ps(fjy0,ty);
 403             fjz0             = _mm256_add_ps(fjz0,tz);
 404
 405             }
 406
 407             fjptrA             = f+j_coord_offsetA;
 408             fjptrB             = f+j_coord_offsetB;
 409             fjptrC             = f+j_coord_offsetC;
 410             fjptrD             = f+j_coord_offsetD;
 411             fjptrE             = f+j_coord_offsetE;
 412             fjptrF             = f+j_coord_offsetF;
 413             fjptrG             = f+j_coord_offsetG;
 414             fjptrH             = f+j_coord_offsetH;
 415
 416             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 330 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             jnrlistE         = jjnr[jidx+4];
 430             jnrlistF         = jjnr[jidx+5];
 431             jnrlistG         = jjnr[jidx+6];
 432             jnrlistH         = jjnr[jidx+7];
 433             /* Sign of each element will be negative for non-real atoms.
 434              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 435              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 436              */
 437             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 438                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 439
 440             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 441             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 442             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 443             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 444             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 445             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 446             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 447             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 448             j_coord_offsetA  = DIM*jnrA;
 449             j_coord_offsetB  = DIM*jnrB;
 450             j_coord_offsetC  = DIM*jnrC;
 451             j_coord_offsetD  = DIM*jnrD;
 452             j_coord_offsetE  = DIM*jnrE;
 453             j_coord_offsetF  = DIM*jnrF;
 454             j_coord_offsetG  = DIM*jnrG;
 455             j_coord_offsetH  = DIM*jnrH;
 456
 457             /* load j atom coordinates */
 458             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 460                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 461                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 462                                                  &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm256_sub_ps(ix0,jx0);
 466             dy00             = _mm256_sub_ps(iy0,jy0);
 467             dz00             = _mm256_sub_ps(iz0,jz0);
 468             dx10             = _mm256_sub_ps(ix1,jx0);
 469             dy10             = _mm256_sub_ps(iy1,jy0);
 470             dz10             = _mm256_sub_ps(iz1,jz0);
 471             dx20             = _mm256_sub_ps(ix2,jx0);
 472             dy20             = _mm256_sub_ps(iy2,jy0);
 473             dz20             = _mm256_sub_ps(iz2,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 477             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 478             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 479
 480             rinv00           = avx256_invsqrt_f(rsq00);
 481             rinv10           = avx256_invsqrt_f(rsq10);
 482             rinv20           = avx256_invsqrt_f(rsq20);
 483
 484             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 485             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 486             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 490                                                                  charge+jnrC+0,charge+jnrD+0,
 491                                                                  charge+jnrE+0,charge+jnrF+0,
 492                                                                  charge+jnrG+0,charge+jnrH+0);
 493
 494             fjx0             = _mm256_setzero_ps();
 495             fjy0             = _mm256_setzero_ps();
 496             fjz0             = _mm256_setzero_ps();
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 503             {
 504
 505             r00              = _mm256_mul_ps(rsq00,rinv00);
 506             r00              = _mm256_andnot_ps(dummy_mask,r00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm256_mul_ps(iq0,jq0);
 510
 511             /* EWALD ELECTROSTATICS */
 512
 513             /* Analytical PME correction */
 514             zeta2            = _mm256_mul_ps(beta2,rsq00);
 515             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 516             pmecorrF         = avx256_pmecorrF_f(zeta2);
 517             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 518             felec            = _mm256_mul_ps(qq00,felec);
 519             pmecorrV         = avx256_pmecorrV_f(zeta2);
 520             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 521             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 522             velec            = _mm256_mul_ps(qq00,velec);
 523
 524             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velec            = _mm256_and_ps(velec,cutoff_mask);
 528             velec            = _mm256_andnot_ps(dummy_mask,velec);
 529             velecsum         = _mm256_add_ps(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 534
 535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx00);
 539             ty               = _mm256_mul_ps(fscal,dy00);
 540             tz               = _mm256_mul_ps(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm256_add_ps(fix0,tx);
 544             fiy0             = _mm256_add_ps(fiy0,ty);
 545             fiz0             = _mm256_add_ps(fiz0,tz);
 546
 547             fjx0             = _mm256_add_ps(fjx0,tx);
 548             fjy0             = _mm256_add_ps(fjy0,ty);
 549             fjz0             = _mm256_add_ps(fjz0,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 558             {
 559
 560             r10              = _mm256_mul_ps(rsq10,rinv10);
 561             r10              = _mm256_andnot_ps(dummy_mask,r10);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq10             = _mm256_mul_ps(iq1,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm256_mul_ps(beta2,rsq10);
 570             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 571             pmecorrF         = avx256_pmecorrF_f(zeta2);
 572             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 573             felec            = _mm256_mul_ps(qq10,felec);
 574             pmecorrV         = avx256_pmecorrV_f(zeta2);
 575             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 576             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 577             velec            = _mm256_mul_ps(qq10,velec);
 578
 579             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm256_and_ps(velec,cutoff_mask);
 583             velec            = _mm256_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm256_add_ps(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 589
 590             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm256_mul_ps(fscal,dx10);
 594             ty               = _mm256_mul_ps(fscal,dy10);
 595             tz               = _mm256_mul_ps(fscal,dz10);
 596
 597             /* Update vectorial force */
 598             fix1             = _mm256_add_ps(fix1,tx);
 599             fiy1             = _mm256_add_ps(fiy1,ty);
 600             fiz1             = _mm256_add_ps(fiz1,tz);
 601
 602             fjx0             = _mm256_add_ps(fjx0,tx);
 603             fjy0             = _mm256_add_ps(fjy0,ty);
 604             fjz0             = _mm256_add_ps(fjz0,tz);
 605
 606             }
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 613             {
 614
 615             r20              = _mm256_mul_ps(rsq20,rinv20);
 616             r20              = _mm256_andnot_ps(dummy_mask,r20);
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq20             = _mm256_mul_ps(iq2,jq0);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Analytical PME correction */
 624             zeta2            = _mm256_mul_ps(beta2,rsq20);
 625             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 626             pmecorrF         = avx256_pmecorrF_f(zeta2);
 627             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 628             felec            = _mm256_mul_ps(qq20,felec);
 629             pmecorrV         = avx256_pmecorrV_f(zeta2);
 630             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 631             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 632             velec            = _mm256_mul_ps(qq20,velec);
 633
 634             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm256_and_ps(velec,cutoff_mask);
 638             velec            = _mm256_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm256_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 644
 645             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_ps(fscal,dx20);
 649             ty               = _mm256_mul_ps(fscal,dy20);
 650             tz               = _mm256_mul_ps(fscal,dz20);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm256_add_ps(fix2,tx);
 654             fiy2             = _mm256_add_ps(fiy2,ty);
 655             fiz2             = _mm256_add_ps(fiz2,tz);
 656
 657             fjx0             = _mm256_add_ps(fjx0,tx);
 658             fjy0             = _mm256_add_ps(fjy0,ty);
 659             fjz0             = _mm256_add_ps(fjz0,tz);
 660
 661             }
 662
 663             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 664             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 665             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 666             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 667             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 668             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 669             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 670             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 671
 672             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 673
 674             /* Inner loop uses 333 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 680                                                  f+i_coord_offset,fshift+i_shift_offset);
 681
 682         ggid                        = gid[iidx];
 683         /* Update potential energies */
 684         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 685
 686         /* Increment number of inner iterations */
 687         inneriter                  += j_index_end - j_index_start;
 688
 689         /* Outer loop uses 19 flops */
 690     }
 691
 692     /* Increment number of outer iterations */
 693     outeriter        += nri;
 694
 695     /* Update outer/inner flops */
 696
 697     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*333);
 698 }
 699 /*
 700  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 701  * Electrostatics interaction: Ewald
 702  * VdW interaction:            None
 703  * Geometry:                   Water3-Particle
 704  * Calculate force/pot:        Force
 705  */
 706 void
 707 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 708                     (t_nblist                    * gmx_restrict       nlist,
 709                      rvec                        * gmx_restrict          xx,
 710                      rvec                        * gmx_restrict          ff,
 711                      struct t_forcerec           * gmx_restrict          fr,
 712                      t_mdatoms                   * gmx_restrict     mdatoms,
 713                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 714                      t_nrnb                      * gmx_restrict        nrnb)
 715 {
 716     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 717      * just 0 for non-waters.
 718      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 719      * jnr indices corresponding to data put in the four positions in the SIMD register.
 720      */
 721     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 722     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 723     int              jnrA,jnrB,jnrC,jnrD;
 724     int              jnrE,jnrF,jnrG,jnrH;
 725     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 726     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 727     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 728     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 729     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 730     real             rcutoff_scalar;
 731     real             *shiftvec,*fshift,*x,*f;
 732     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 733     real             scratch[4*DIM];
 734     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 735     real *           vdwioffsetptr0;
 736     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 737     real *           vdwioffsetptr1;
 738     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 739     real *           vdwioffsetptr2;
 740     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 741     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 742     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 743     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 744     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 745     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 746     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 747     real             *charge;
 748     __m256i          ewitab;
 749     __m128i          ewitab_lo,ewitab_hi;
 750     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 751     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 752     real             *ewtab;
 753     __m256           dummy_mask,cutoff_mask;
 754     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 755     __m256           one     = _mm256_set1_ps(1.0);
 756     __m256           two     = _mm256_set1_ps(2.0);
 757     x                = xx[0];
 758     f                = ff[0];
 759
 760     nri              = nlist->nri;
 761     iinr             = nlist->iinr;
 762     jindex           = nlist->jindex;
 763     jjnr             = nlist->jjnr;
 764     shiftidx         = nlist->shift;
 765     gid              = nlist->gid;
 766     shiftvec         = fr->shift_vec[0];
 767     fshift           = fr->fshift[0];
 768     facel            = _mm256_set1_ps(fr->ic->epsfac);
 769     charge           = mdatoms->chargeA;
 770
 771     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 772     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 773     beta2            = _mm256_mul_ps(beta,beta);
 774     beta3            = _mm256_mul_ps(beta,beta2);
 775
 776     ewtab            = fr->ic->tabq_coul_F;
 777     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 778     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 783     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 784     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 785
 786     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 787     rcutoff_scalar   = fr->ic->rcoulomb;
 788     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 789     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 790
 791     /* Avoid stupid compiler warnings */
 792     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 793     j_coord_offsetA = 0;
 794     j_coord_offsetB = 0;
 795     j_coord_offsetC = 0;
 796     j_coord_offsetD = 0;
 797     j_coord_offsetE = 0;
 798     j_coord_offsetF = 0;
 799     j_coord_offsetG = 0;
 800     j_coord_offsetH = 0;
 801
 802     outeriter        = 0;
 803     inneriter        = 0;
 804
 805     for(iidx=0;iidx<4*DIM;iidx++)
 806     {
 807         scratch[iidx] = 0.0;
 808     }
 809
 810     /* Start outer loop over neighborlists */
 811     for(iidx=0; iidx<nri; iidx++)
 812     {
 813         /* Load shift vector for this list */
 814         i_shift_offset   = DIM*shiftidx[iidx];
 815
 816         /* Load limits for loop over neighbors */
 817         j_index_start    = jindex[iidx];
 818         j_index_end      = jindex[iidx+1];
 819
 820         /* Get outer coordinate index */
 821         inr              = iinr[iidx];
 822         i_coord_offset   = DIM*inr;
 823
 824         /* Load i particle coords and add shift vector */
 825         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 826                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 827
 828         fix0             = _mm256_setzero_ps();
 829         fiy0             = _mm256_setzero_ps();
 830         fiz0             = _mm256_setzero_ps();
 831         fix1             = _mm256_setzero_ps();
 832         fiy1             = _mm256_setzero_ps();
 833         fiz1             = _mm256_setzero_ps();
 834         fix2             = _mm256_setzero_ps();
 835         fiy2             = _mm256_setzero_ps();
 836         fiz2             = _mm256_setzero_ps();
 837
 838         /* Start inner kernel loop */
 839         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 840         {
 841
 842             /* Get j neighbor index, and coordinate index */
 843             jnrA             = jjnr[jidx];
 844             jnrB             = jjnr[jidx+1];
 845             jnrC             = jjnr[jidx+2];
 846             jnrD             = jjnr[jidx+3];
 847             jnrE             = jjnr[jidx+4];
 848             jnrF             = jjnr[jidx+5];
 849             jnrG             = jjnr[jidx+6];
 850             jnrH             = jjnr[jidx+7];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855             j_coord_offsetE  = DIM*jnrE;
 856             j_coord_offsetF  = DIM*jnrF;
 857             j_coord_offsetG  = DIM*jnrG;
 858             j_coord_offsetH  = DIM*jnrH;
 859
 860             /* load j atom coordinates */
 861             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 862                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 863                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 864                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 865                                                  &jx0,&jy0,&jz0);
 866
 867             /* Calculate displacement vector */
 868             dx00             = _mm256_sub_ps(ix0,jx0);
 869             dy00             = _mm256_sub_ps(iy0,jy0);
 870             dz00             = _mm256_sub_ps(iz0,jz0);
 871             dx10             = _mm256_sub_ps(ix1,jx0);
 872             dy10             = _mm256_sub_ps(iy1,jy0);
 873             dz10             = _mm256_sub_ps(iz1,jz0);
 874             dx20             = _mm256_sub_ps(ix2,jx0);
 875             dy20             = _mm256_sub_ps(iy2,jy0);
 876             dz20             = _mm256_sub_ps(iz2,jz0);
 877
 878             /* Calculate squared distance and things based on it */
 879             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 880             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 881             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 882
 883             rinv00           = avx256_invsqrt_f(rsq00);
 884             rinv10           = avx256_invsqrt_f(rsq10);
 885             rinv20           = avx256_invsqrt_f(rsq20);
 886
 887             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 888             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 889             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 890
 891             /* Load parameters for j particles */
 892             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 893                                                                  charge+jnrC+0,charge+jnrD+0,
 894                                                                  charge+jnrE+0,charge+jnrF+0,
 895                                                                  charge+jnrG+0,charge+jnrH+0);
 896
 897             fjx0             = _mm256_setzero_ps();
 898             fjy0             = _mm256_setzero_ps();
 899             fjz0             = _mm256_setzero_ps();
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 906             {
 907
 908             r00              = _mm256_mul_ps(rsq00,rinv00);
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq00             = _mm256_mul_ps(iq0,jq0);
 912
 913             /* EWALD ELECTROSTATICS */
 914
 915             /* Analytical PME correction */
 916             zeta2            = _mm256_mul_ps(beta2,rsq00);
 917             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 918             pmecorrF         = avx256_pmecorrF_f(zeta2);
 919             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 920             felec            = _mm256_mul_ps(qq00,felec);
 921
 922             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 923
 924             fscal            = felec;
 925
 926             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm256_mul_ps(fscal,dx00);
 930             ty               = _mm256_mul_ps(fscal,dy00);
 931             tz               = _mm256_mul_ps(fscal,dz00);
 932
 933             /* Update vectorial force */
 934             fix0             = _mm256_add_ps(fix0,tx);
 935             fiy0             = _mm256_add_ps(fiy0,ty);
 936             fiz0             = _mm256_add_ps(fiz0,tz);
 937
 938             fjx0             = _mm256_add_ps(fjx0,tx);
 939             fjy0             = _mm256_add_ps(fjy0,ty);
 940             fjz0             = _mm256_add_ps(fjz0,tz);
 941
 942             }
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 949             {
 950
 951             r10              = _mm256_mul_ps(rsq10,rinv10);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm256_mul_ps(iq1,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Analytical PME correction */
 959             zeta2            = _mm256_mul_ps(beta2,rsq10);
 960             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 961             pmecorrF         = avx256_pmecorrF_f(zeta2);
 962             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 963             felec            = _mm256_mul_ps(qq10,felec);
 964
 965             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 966
 967             fscal            = felec;
 968
 969             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm256_mul_ps(fscal,dx10);
 973             ty               = _mm256_mul_ps(fscal,dy10);
 974             tz               = _mm256_mul_ps(fscal,dz10);
 975
 976             /* Update vectorial force */
 977             fix1             = _mm256_add_ps(fix1,tx);
 978             fiy1             = _mm256_add_ps(fiy1,ty);
 979             fiz1             = _mm256_add_ps(fiz1,tz);
 980
 981             fjx0             = _mm256_add_ps(fjx0,tx);
 982             fjy0             = _mm256_add_ps(fjy0,ty);
 983             fjz0             = _mm256_add_ps(fjz0,tz);
 984
 985             }
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 992             {
 993
 994             r20              = _mm256_mul_ps(rsq20,rinv20);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq20             = _mm256_mul_ps(iq2,jq0);
 998
 999             /* EWALD ELECTROSTATICS */
1000
1001             /* Analytical PME correction */
1002             zeta2            = _mm256_mul_ps(beta2,rsq20);
1003             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1004             pmecorrF         = avx256_pmecorrF_f(zeta2);
1005             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1006             felec            = _mm256_mul_ps(qq20,felec);
1007
1008             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx20);
1016             ty               = _mm256_mul_ps(fscal,dy20);
1017             tz               = _mm256_mul_ps(fscal,dz20);
1018
1019             /* Update vectorial force */
1020             fix2             = _mm256_add_ps(fix2,tx);
1021             fiy2             = _mm256_add_ps(fiy2,ty);
1022             fiz2             = _mm256_add_ps(fiz2,tz);
1023
1024             fjx0             = _mm256_add_ps(fjx0,tx);
1025             fjy0             = _mm256_add_ps(fjy0,ty);
1026             fjz0             = _mm256_add_ps(fjz0,tz);
1027
1028             }
1029
1030             fjptrA             = f+j_coord_offsetA;
1031             fjptrB             = f+j_coord_offsetB;
1032             fjptrC             = f+j_coord_offsetC;
1033             fjptrD             = f+j_coord_offsetD;
1034             fjptrE             = f+j_coord_offsetE;
1035             fjptrF             = f+j_coord_offsetF;
1036             fjptrG             = f+j_coord_offsetG;
1037             fjptrH             = f+j_coord_offsetH;
1038
1039             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1040
1041             /* Inner loop uses 180 flops */
1042         }
1043
1044         if(jidx<j_index_end)
1045         {
1046
1047             /* Get j neighbor index, and coordinate index */
1048             jnrlistA         = jjnr[jidx];
1049             jnrlistB         = jjnr[jidx+1];
1050             jnrlistC         = jjnr[jidx+2];
1051             jnrlistD         = jjnr[jidx+3];
1052             jnrlistE         = jjnr[jidx+4];
1053             jnrlistF         = jjnr[jidx+5];
1054             jnrlistG         = jjnr[jidx+6];
1055             jnrlistH         = jjnr[jidx+7];
1056             /* Sign of each element will be negative for non-real atoms.
1057              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1058              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1059              */
1060             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1061                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1062
1063             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1064             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1065             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1066             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1067             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1068             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1069             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1070             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1071             j_coord_offsetA  = DIM*jnrA;
1072             j_coord_offsetB  = DIM*jnrB;
1073             j_coord_offsetC  = DIM*jnrC;
1074             j_coord_offsetD  = DIM*jnrD;
1075             j_coord_offsetE  = DIM*jnrE;
1076             j_coord_offsetF  = DIM*jnrF;
1077             j_coord_offsetG  = DIM*jnrG;
1078             j_coord_offsetH  = DIM*jnrH;
1079
1080             /* load j atom coordinates */
1081             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1082                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1083                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1084                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1085                                                  &jx0,&jy0,&jz0);
1086
1087             /* Calculate displacement vector */
1088             dx00             = _mm256_sub_ps(ix0,jx0);
1089             dy00             = _mm256_sub_ps(iy0,jy0);
1090             dz00             = _mm256_sub_ps(iz0,jz0);
1091             dx10             = _mm256_sub_ps(ix1,jx0);
1092             dy10             = _mm256_sub_ps(iy1,jy0);
1093             dz10             = _mm256_sub_ps(iz1,jz0);
1094             dx20             = _mm256_sub_ps(ix2,jx0);
1095             dy20             = _mm256_sub_ps(iy2,jy0);
1096             dz20             = _mm256_sub_ps(iz2,jz0);
1097
1098             /* Calculate squared distance and things based on it */
1099             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1100             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1101             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1102
1103             rinv00           = avx256_invsqrt_f(rsq00);
1104             rinv10           = avx256_invsqrt_f(rsq10);
1105             rinv20           = avx256_invsqrt_f(rsq20);
1106
1107             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1108             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1109             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1110
1111             /* Load parameters for j particles */
1112             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1113                                                                  charge+jnrC+0,charge+jnrD+0,
1114                                                                  charge+jnrE+0,charge+jnrF+0,
1115                                                                  charge+jnrG+0,charge+jnrH+0);
1116
1117             fjx0             = _mm256_setzero_ps();
1118             fjy0             = _mm256_setzero_ps();
1119             fjz0             = _mm256_setzero_ps();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1126             {
1127
1128             r00              = _mm256_mul_ps(rsq00,rinv00);
1129             r00              = _mm256_andnot_ps(dummy_mask,r00);
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq00             = _mm256_mul_ps(iq0,jq0);
1133
1134             /* EWALD ELECTROSTATICS */
1135
1136             /* Analytical PME correction */
1137             zeta2            = _mm256_mul_ps(beta2,rsq00);
1138             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1139             pmecorrF         = avx256_pmecorrF_f(zeta2);
1140             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1141             felec            = _mm256_mul_ps(qq00,felec);
1142
1143             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1148
1149             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1150
1151             /* Calculate temporary vectorial force */
1152             tx               = _mm256_mul_ps(fscal,dx00);
1153             ty               = _mm256_mul_ps(fscal,dy00);
1154             tz               = _mm256_mul_ps(fscal,dz00);
1155
1156             /* Update vectorial force */
1157             fix0             = _mm256_add_ps(fix0,tx);
1158             fiy0             = _mm256_add_ps(fiy0,ty);
1159             fiz0             = _mm256_add_ps(fiz0,tz);
1160
1161             fjx0             = _mm256_add_ps(fjx0,tx);
1162             fjy0             = _mm256_add_ps(fjy0,ty);
1163             fjz0             = _mm256_add_ps(fjz0,tz);
1164
1165             }
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1172             {
1173
1174             r10              = _mm256_mul_ps(rsq10,rinv10);
1175             r10              = _mm256_andnot_ps(dummy_mask,r10);
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq10             = _mm256_mul_ps(iq1,jq0);
1179
1180             /* EWALD ELECTROSTATICS */
1181
1182             /* Analytical PME correction */
1183             zeta2            = _mm256_mul_ps(beta2,rsq10);
1184             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1185             pmecorrF         = avx256_pmecorrF_f(zeta2);
1186             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1187             felec            = _mm256_mul_ps(qq10,felec);
1188
1189             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1190
1191             fscal            = felec;
1192
1193             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1194
1195             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1196
1197             /* Calculate temporary vectorial force */
1198             tx               = _mm256_mul_ps(fscal,dx10);
1199             ty               = _mm256_mul_ps(fscal,dy10);
1200             tz               = _mm256_mul_ps(fscal,dz10);
1201
1202             /* Update vectorial force */
1203             fix1             = _mm256_add_ps(fix1,tx);
1204             fiy1             = _mm256_add_ps(fiy1,ty);
1205             fiz1             = _mm256_add_ps(fiz1,tz);
1206
1207             fjx0             = _mm256_add_ps(fjx0,tx);
1208             fjy0             = _mm256_add_ps(fjy0,ty);
1209             fjz0             = _mm256_add_ps(fjz0,tz);
1210
1211             }
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1218             {
1219
1220             r20              = _mm256_mul_ps(rsq20,rinv20);
1221             r20              = _mm256_andnot_ps(dummy_mask,r20);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             qq20             = _mm256_mul_ps(iq2,jq0);
1225
1226             /* EWALD ELECTROSTATICS */
1227
1228             /* Analytical PME correction */
1229             zeta2            = _mm256_mul_ps(beta2,rsq20);
1230             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1231             pmecorrF         = avx256_pmecorrF_f(zeta2);
1232             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1233             felec            = _mm256_mul_ps(qq20,felec);
1234
1235             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1240
1241             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1242
1243             /* Calculate temporary vectorial force */
1244             tx               = _mm256_mul_ps(fscal,dx20);
1245             ty               = _mm256_mul_ps(fscal,dy20);
1246             tz               = _mm256_mul_ps(fscal,dz20);
1247
1248             /* Update vectorial force */
1249             fix2             = _mm256_add_ps(fix2,tx);
1250             fiy2             = _mm256_add_ps(fiy2,ty);
1251             fiz2             = _mm256_add_ps(fiz2,tz);
1252
1253             fjx0             = _mm256_add_ps(fjx0,tx);
1254             fjy0             = _mm256_add_ps(fjy0,ty);
1255             fjz0             = _mm256_add_ps(fjz0,tz);
1256
1257             }
1258
1259             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1260             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1261             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1262             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1263             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1264             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1265             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1266             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1267
1268             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1269
1270             /* Inner loop uses 183 flops */
1271         }
1272
1273         /* End of innermost loop */
1274
1275         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1276                                                  f+i_coord_offset,fshift+i_shift_offset);
1277
1278         /* Increment number of inner iterations */
1279         inneriter                  += j_index_end - j_index_start;
1280
1281         /* Outer loop uses 18 flops */
1282     }
1283
1284     /* Increment number of outer iterations */
1285     outeriter        += nri;
1286
1287     /* Update outer/inner flops */
1288
1289     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*183);
1290 }