src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecEw_VdwNone_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  86     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     real *           vdwioffsetptr3;
  90     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 133     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 134     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 171                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix1             = _mm256_setzero_ps();
 174         fiy1             = _mm256_setzero_ps();
 175         fiz1             = _mm256_setzero_ps();
 176         fix2             = _mm256_setzero_ps();
 177         fiy2             = _mm256_setzero_ps();
 178         fiz2             = _mm256_setzero_ps();
 179         fix3             = _mm256_setzero_ps();
 180         fiy3             = _mm256_setzero_ps();
 181         fiz3             = _mm256_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             jnrE             = jjnr[jidx+4];
 196             jnrF             = jjnr[jidx+5];
 197             jnrG             = jjnr[jidx+6];
 198             jnrH             = jjnr[jidx+7];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203             j_coord_offsetE  = DIM*jnrE;
 204             j_coord_offsetF  = DIM*jnrF;
 205             j_coord_offsetG  = DIM*jnrG;
 206             j_coord_offsetH  = DIM*jnrH;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 212                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx10             = _mm256_sub_ps(ix1,jx0);
 217             dy10             = _mm256_sub_ps(iy1,jy0);
 218             dz10             = _mm256_sub_ps(iz1,jz0);
 219             dx20             = _mm256_sub_ps(ix2,jx0);
 220             dy20             = _mm256_sub_ps(iy2,jy0);
 221             dz20             = _mm256_sub_ps(iz2,jz0);
 222             dx30             = _mm256_sub_ps(ix3,jx0);
 223             dy30             = _mm256_sub_ps(iy3,jy0);
 224             dz30             = _mm256_sub_ps(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv10           = avx256_invsqrt_f(rsq10);
 232             rinv20           = avx256_invsqrt_f(rsq20);
 233             rinv30           = avx256_invsqrt_f(rsq30);
 234
 235             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 236             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 237             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0,
 242                                                                  charge+jnrE+0,charge+jnrF+0,
 243                                                                  charge+jnrG+0,charge+jnrH+0);
 244
 245             fjx0             = _mm256_setzero_ps();
 246             fjy0             = _mm256_setzero_ps();
 247             fjz0             = _mm256_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r10              = _mm256_mul_ps(rsq10,rinv10);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq10             = _mm256_mul_ps(iq1,jq0);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Analytical PME correction */
 261             zeta2            = _mm256_mul_ps(beta2,rsq10);
 262             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 263             pmecorrF         = avx256_pmecorrF_f(zeta2);
 264             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 265             felec            = _mm256_mul_ps(qq10,felec);
 266             pmecorrV         = avx256_pmecorrV_f(zeta2);
 267             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 268             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 269             velec            = _mm256_mul_ps(qq10,velec);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             velecsum         = _mm256_add_ps(velecsum,velec);
 273
 274             fscal            = felec;
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm256_mul_ps(fscal,dx10);
 278             ty               = _mm256_mul_ps(fscal,dy10);
 279             tz               = _mm256_mul_ps(fscal,dz10);
 280
 281             /* Update vectorial force */
 282             fix1             = _mm256_add_ps(fix1,tx);
 283             fiy1             = _mm256_add_ps(fiy1,ty);
 284             fiz1             = _mm256_add_ps(fiz1,tz);
 285
 286             fjx0             = _mm256_add_ps(fjx0,tx);
 287             fjy0             = _mm256_add_ps(fjy0,ty);
 288             fjz0             = _mm256_add_ps(fjz0,tz);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r20              = _mm256_mul_ps(rsq20,rinv20);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq20             = _mm256_mul_ps(iq2,jq0);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Analytical PME correction */
 302             zeta2            = _mm256_mul_ps(beta2,rsq20);
 303             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 304             pmecorrF         = avx256_pmecorrF_f(zeta2);
 305             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 306             felec            = _mm256_mul_ps(qq20,felec);
 307             pmecorrV         = avx256_pmecorrV_f(zeta2);
 308             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 309             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 310             velec            = _mm256_mul_ps(qq20,velec);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm256_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_ps(fscal,dx20);
 319             ty               = _mm256_mul_ps(fscal,dy20);
 320             tz               = _mm256_mul_ps(fscal,dz20);
 321
 322             /* Update vectorial force */
 323             fix2             = _mm256_add_ps(fix2,tx);
 324             fiy2             = _mm256_add_ps(fiy2,ty);
 325             fiz2             = _mm256_add_ps(fiz2,tz);
 326
 327             fjx0             = _mm256_add_ps(fjx0,tx);
 328             fjy0             = _mm256_add_ps(fjy0,ty);
 329             fjz0             = _mm256_add_ps(fjz0,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r30              = _mm256_mul_ps(rsq30,rinv30);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq30             = _mm256_mul_ps(iq3,jq0);
 339
 340             /* EWALD ELECTROSTATICS */
 341
 342             /* Analytical PME correction */
 343             zeta2            = _mm256_mul_ps(beta2,rsq30);
 344             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 345             pmecorrF         = avx256_pmecorrF_f(zeta2);
 346             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 347             felec            = _mm256_mul_ps(qq30,felec);
 348             pmecorrV         = avx256_pmecorrV_f(zeta2);
 349             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 350             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 351             velec            = _mm256_mul_ps(qq30,velec);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velecsum         = _mm256_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm256_mul_ps(fscal,dx30);
 360             ty               = _mm256_mul_ps(fscal,dy30);
 361             tz               = _mm256_mul_ps(fscal,dz30);
 362
 363             /* Update vectorial force */
 364             fix3             = _mm256_add_ps(fix3,tx);
 365             fiy3             = _mm256_add_ps(fiy3,ty);
 366             fiz3             = _mm256_add_ps(fiz3,tz);
 367
 368             fjx0             = _mm256_add_ps(fjx0,tx);
 369             fjy0             = _mm256_add_ps(fjy0,ty);
 370             fjz0             = _mm256_add_ps(fjz0,tz);
 371
 372             fjptrA             = f+j_coord_offsetA;
 373             fjptrB             = f+j_coord_offsetB;
 374             fjptrC             = f+j_coord_offsetC;
 375             fjptrD             = f+j_coord_offsetD;
 376             fjptrE             = f+j_coord_offsetE;
 377             fjptrF             = f+j_coord_offsetF;
 378             fjptrG             = f+j_coord_offsetG;
 379             fjptrH             = f+j_coord_offsetH;
 380
 381             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 255 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             /* Get j neighbor index, and coordinate index */
 390             jnrlistA         = jjnr[jidx];
 391             jnrlistB         = jjnr[jidx+1];
 392             jnrlistC         = jjnr[jidx+2];
 393             jnrlistD         = jjnr[jidx+3];
 394             jnrlistE         = jjnr[jidx+4];
 395             jnrlistF         = jjnr[jidx+5];
 396             jnrlistG         = jjnr[jidx+6];
 397             jnrlistH         = jjnr[jidx+7];
 398             /* Sign of each element will be negative for non-real atoms.
 399              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 400              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 401              */
 402             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 403                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 404
 405             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 406             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 407             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 408             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 409             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 410             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 411             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 412             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 413             j_coord_offsetA  = DIM*jnrA;
 414             j_coord_offsetB  = DIM*jnrB;
 415             j_coord_offsetC  = DIM*jnrC;
 416             j_coord_offsetD  = DIM*jnrD;
 417             j_coord_offsetE  = DIM*jnrE;
 418             j_coord_offsetF  = DIM*jnrF;
 419             j_coord_offsetG  = DIM*jnrG;
 420             j_coord_offsetH  = DIM*jnrH;
 421
 422             /* load j atom coordinates */
 423             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 424                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 425                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 426                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 427                                                  &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx10             = _mm256_sub_ps(ix1,jx0);
 431             dy10             = _mm256_sub_ps(iy1,jy0);
 432             dz10             = _mm256_sub_ps(iz1,jz0);
 433             dx20             = _mm256_sub_ps(ix2,jx0);
 434             dy20             = _mm256_sub_ps(iy2,jy0);
 435             dz20             = _mm256_sub_ps(iz2,jz0);
 436             dx30             = _mm256_sub_ps(ix3,jx0);
 437             dy30             = _mm256_sub_ps(iy3,jy0);
 438             dz30             = _mm256_sub_ps(iz3,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 442             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 443             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 444
 445             rinv10           = avx256_invsqrt_f(rsq10);
 446             rinv20           = avx256_invsqrt_f(rsq20);
 447             rinv30           = avx256_invsqrt_f(rsq30);
 448
 449             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 450             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 451             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 452
 453             /* Load parameters for j particles */
 454             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 455                                                                  charge+jnrC+0,charge+jnrD+0,
 456                                                                  charge+jnrE+0,charge+jnrF+0,
 457                                                                  charge+jnrG+0,charge+jnrH+0);
 458
 459             fjx0             = _mm256_setzero_ps();
 460             fjy0             = _mm256_setzero_ps();
 461             fjz0             = _mm256_setzero_ps();
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r10              = _mm256_mul_ps(rsq10,rinv10);
 468             r10              = _mm256_andnot_ps(dummy_mask,r10);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq10             = _mm256_mul_ps(iq1,jq0);
 472
 473             /* EWALD ELECTROSTATICS */
 474
 475             /* Analytical PME correction */
 476             zeta2            = _mm256_mul_ps(beta2,rsq10);
 477             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 478             pmecorrF         = avx256_pmecorrF_f(zeta2);
 479             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 480             felec            = _mm256_mul_ps(qq10,felec);
 481             pmecorrV         = avx256_pmecorrV_f(zeta2);
 482             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 483             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 484             velec            = _mm256_mul_ps(qq10,velec);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm256_andnot_ps(dummy_mask,velec);
 488             velecsum         = _mm256_add_ps(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_ps(fscal,dx10);
 496             ty               = _mm256_mul_ps(fscal,dy10);
 497             tz               = _mm256_mul_ps(fscal,dz10);
 498
 499             /* Update vectorial force */
 500             fix1             = _mm256_add_ps(fix1,tx);
 501             fiy1             = _mm256_add_ps(fiy1,ty);
 502             fiz1             = _mm256_add_ps(fiz1,tz);
 503
 504             fjx0             = _mm256_add_ps(fjx0,tx);
 505             fjy0             = _mm256_add_ps(fjy0,ty);
 506             fjz0             = _mm256_add_ps(fjz0,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r20              = _mm256_mul_ps(rsq20,rinv20);
 513             r20              = _mm256_andnot_ps(dummy_mask,r20);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq20             = _mm256_mul_ps(iq2,jq0);
 517
 518             /* EWALD ELECTROSTATICS */
 519
 520             /* Analytical PME correction */
 521             zeta2            = _mm256_mul_ps(beta2,rsq20);
 522             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 523             pmecorrF         = avx256_pmecorrF_f(zeta2);
 524             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 525             felec            = _mm256_mul_ps(qq20,felec);
 526             pmecorrV         = avx256_pmecorrV_f(zeta2);
 527             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 528             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 529             velec            = _mm256_mul_ps(qq20,velec);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm256_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm256_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm256_mul_ps(fscal,dx20);
 541             ty               = _mm256_mul_ps(fscal,dy20);
 542             tz               = _mm256_mul_ps(fscal,dz20);
 543
 544             /* Update vectorial force */
 545             fix2             = _mm256_add_ps(fix2,tx);
 546             fiy2             = _mm256_add_ps(fiy2,ty);
 547             fiz2             = _mm256_add_ps(fiz2,tz);
 548
 549             fjx0             = _mm256_add_ps(fjx0,tx);
 550             fjy0             = _mm256_add_ps(fjy0,ty);
 551             fjz0             = _mm256_add_ps(fjz0,tz);
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r30              = _mm256_mul_ps(rsq30,rinv30);
 558             r30              = _mm256_andnot_ps(dummy_mask,r30);
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq30             = _mm256_mul_ps(iq3,jq0);
 562
 563             /* EWALD ELECTROSTATICS */
 564
 565             /* Analytical PME correction */
 566             zeta2            = _mm256_mul_ps(beta2,rsq30);
 567             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 568             pmecorrF         = avx256_pmecorrF_f(zeta2);
 569             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 570             felec            = _mm256_mul_ps(qq30,felec);
 571             pmecorrV         = avx256_pmecorrV_f(zeta2);
 572             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 573             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 574             velec            = _mm256_mul_ps(qq30,velec);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm256_andnot_ps(dummy_mask,velec);
 578             velecsum         = _mm256_add_ps(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_ps(fscal,dx30);
 586             ty               = _mm256_mul_ps(fscal,dy30);
 587             tz               = _mm256_mul_ps(fscal,dz30);
 588
 589             /* Update vectorial force */
 590             fix3             = _mm256_add_ps(fix3,tx);
 591             fiy3             = _mm256_add_ps(fiy3,ty);
 592             fiz3             = _mm256_add_ps(fiz3,tz);
 593
 594             fjx0             = _mm256_add_ps(fjx0,tx);
 595             fjy0             = _mm256_add_ps(fjy0,ty);
 596             fjz0             = _mm256_add_ps(fjz0,tz);
 597
 598             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 599             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 600             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 601             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 602             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 603             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 604             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 605             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 606
 607             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 608
 609             /* Inner loop uses 258 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 615                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 616
 617         ggid                        = gid[iidx];
 618         /* Update potential energies */
 619         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 19 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*258);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
 636  * Electrostatics interaction: Ewald
 637  * VdW interaction:            None
 638  * Geometry:                   Water4-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      struct t_forcerec           * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB,jnrC,jnrD;
 659     int              jnrE,jnrF,jnrG,jnrH;
 660     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 661     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 662     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 663     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 664     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 665     real             rcutoff_scalar;
 666     real             *shiftvec,*fshift,*x,*f;
 667     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 668     real             scratch[4*DIM];
 669     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 670     real *           vdwioffsetptr1;
 671     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 672     real *           vdwioffsetptr2;
 673     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 674     real *           vdwioffsetptr3;
 675     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 676     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 677     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 678     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 679     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 680     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 681     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 682     real             *charge;
 683     __m256i          ewitab;
 684     __m128i          ewitab_lo,ewitab_hi;
 685     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 686     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 687     real             *ewtab;
 688     __m256           dummy_mask,cutoff_mask;
 689     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 690     __m256           one     = _mm256_set1_ps(1.0);
 691     __m256           two     = _mm256_set1_ps(2.0);
 692     x                = xx[0];
 693     f                = ff[0];
 694
 695     nri              = nlist->nri;
 696     iinr             = nlist->iinr;
 697     jindex           = nlist->jindex;
 698     jjnr             = nlist->jjnr;
 699     shiftidx         = nlist->shift;
 700     gid              = nlist->gid;
 701     shiftvec         = fr->shift_vec[0];
 702     fshift           = fr->fshift[0];
 703     facel            = _mm256_set1_ps(fr->ic->epsfac);
 704     charge           = mdatoms->chargeA;
 705
 706     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 707     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 708     beta2            = _mm256_mul_ps(beta,beta);
 709     beta3            = _mm256_mul_ps(beta,beta2);
 710
 711     ewtab            = fr->ic->tabq_coul_F;
 712     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 713     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 714
 715     /* Setup water-specific parameters */
 716     inr              = nlist->iinr[0];
 717     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 718     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 719     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 720
 721     /* Avoid stupid compiler warnings */
 722     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 723     j_coord_offsetA = 0;
 724     j_coord_offsetB = 0;
 725     j_coord_offsetC = 0;
 726     j_coord_offsetD = 0;
 727     j_coord_offsetE = 0;
 728     j_coord_offsetF = 0;
 729     j_coord_offsetG = 0;
 730     j_coord_offsetH = 0;
 731
 732     outeriter        = 0;
 733     inneriter        = 0;
 734
 735     for(iidx=0;iidx<4*DIM;iidx++)
 736     {
 737         scratch[iidx] = 0.0;
 738     }
 739
 740     /* Start outer loop over neighborlists */
 741     for(iidx=0; iidx<nri; iidx++)
 742     {
 743         /* Load shift vector for this list */
 744         i_shift_offset   = DIM*shiftidx[iidx];
 745
 746         /* Load limits for loop over neighbors */
 747         j_index_start    = jindex[iidx];
 748         j_index_end      = jindex[iidx+1];
 749
 750         /* Get outer coordinate index */
 751         inr              = iinr[iidx];
 752         i_coord_offset   = DIM*inr;
 753
 754         /* Load i particle coords and add shift vector */
 755         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 756                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 757
 758         fix1             = _mm256_setzero_ps();
 759         fiy1             = _mm256_setzero_ps();
 760         fiz1             = _mm256_setzero_ps();
 761         fix2             = _mm256_setzero_ps();
 762         fiy2             = _mm256_setzero_ps();
 763         fiz2             = _mm256_setzero_ps();
 764         fix3             = _mm256_setzero_ps();
 765         fiy3             = _mm256_setzero_ps();
 766         fiz3             = _mm256_setzero_ps();
 767
 768         /* Start inner kernel loop */
 769         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 770         {
 771
 772             /* Get j neighbor index, and coordinate index */
 773             jnrA             = jjnr[jidx];
 774             jnrB             = jjnr[jidx+1];
 775             jnrC             = jjnr[jidx+2];
 776             jnrD             = jjnr[jidx+3];
 777             jnrE             = jjnr[jidx+4];
 778             jnrF             = jjnr[jidx+5];
 779             jnrG             = jjnr[jidx+6];
 780             jnrH             = jjnr[jidx+7];
 781             j_coord_offsetA  = DIM*jnrA;
 782             j_coord_offsetB  = DIM*jnrB;
 783             j_coord_offsetC  = DIM*jnrC;
 784             j_coord_offsetD  = DIM*jnrD;
 785             j_coord_offsetE  = DIM*jnrE;
 786             j_coord_offsetF  = DIM*jnrF;
 787             j_coord_offsetG  = DIM*jnrG;
 788             j_coord_offsetH  = DIM*jnrH;
 789
 790             /* load j atom coordinates */
 791             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 792                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 793                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 794                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 795                                                  &jx0,&jy0,&jz0);
 796
 797             /* Calculate displacement vector */
 798             dx10             = _mm256_sub_ps(ix1,jx0);
 799             dy10             = _mm256_sub_ps(iy1,jy0);
 800             dz10             = _mm256_sub_ps(iz1,jz0);
 801             dx20             = _mm256_sub_ps(ix2,jx0);
 802             dy20             = _mm256_sub_ps(iy2,jy0);
 803             dz20             = _mm256_sub_ps(iz2,jz0);
 804             dx30             = _mm256_sub_ps(ix3,jx0);
 805             dy30             = _mm256_sub_ps(iy3,jy0);
 806             dz30             = _mm256_sub_ps(iz3,jz0);
 807
 808             /* Calculate squared distance and things based on it */
 809             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 810             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 811             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 812
 813             rinv10           = avx256_invsqrt_f(rsq10);
 814             rinv20           = avx256_invsqrt_f(rsq20);
 815             rinv30           = avx256_invsqrt_f(rsq30);
 816
 817             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 818             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 819             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 820
 821             /* Load parameters for j particles */
 822             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 823                                                                  charge+jnrC+0,charge+jnrD+0,
 824                                                                  charge+jnrE+0,charge+jnrF+0,
 825                                                                  charge+jnrG+0,charge+jnrH+0);
 826
 827             fjx0             = _mm256_setzero_ps();
 828             fjy0             = _mm256_setzero_ps();
 829             fjz0             = _mm256_setzero_ps();
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             r10              = _mm256_mul_ps(rsq10,rinv10);
 836
 837             /* Compute parameters for interactions between i and j atoms */
 838             qq10             = _mm256_mul_ps(iq1,jq0);
 839
 840             /* EWALD ELECTROSTATICS */
 841
 842             /* Analytical PME correction */
 843             zeta2            = _mm256_mul_ps(beta2,rsq10);
 844             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 845             pmecorrF         = avx256_pmecorrF_f(zeta2);
 846             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 847             felec            = _mm256_mul_ps(qq10,felec);
 848
 849             fscal            = felec;
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_ps(fscal,dx10);
 853             ty               = _mm256_mul_ps(fscal,dy10);
 854             tz               = _mm256_mul_ps(fscal,dz10);
 855
 856             /* Update vectorial force */
 857             fix1             = _mm256_add_ps(fix1,tx);
 858             fiy1             = _mm256_add_ps(fiy1,ty);
 859             fiz1             = _mm256_add_ps(fiz1,tz);
 860
 861             fjx0             = _mm256_add_ps(fjx0,tx);
 862             fjy0             = _mm256_add_ps(fjy0,ty);
 863             fjz0             = _mm256_add_ps(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r20              = _mm256_mul_ps(rsq20,rinv20);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq20             = _mm256_mul_ps(iq2,jq0);
 873
 874             /* EWALD ELECTROSTATICS */
 875
 876             /* Analytical PME correction */
 877             zeta2            = _mm256_mul_ps(beta2,rsq20);
 878             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 879             pmecorrF         = avx256_pmecorrF_f(zeta2);
 880             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 881             felec            = _mm256_mul_ps(qq20,felec);
 882
 883             fscal            = felec;
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm256_mul_ps(fscal,dx20);
 887             ty               = _mm256_mul_ps(fscal,dy20);
 888             tz               = _mm256_mul_ps(fscal,dz20);
 889
 890             /* Update vectorial force */
 891             fix2             = _mm256_add_ps(fix2,tx);
 892             fiy2             = _mm256_add_ps(fiy2,ty);
 893             fiz2             = _mm256_add_ps(fiz2,tz);
 894
 895             fjx0             = _mm256_add_ps(fjx0,tx);
 896             fjy0             = _mm256_add_ps(fjy0,ty);
 897             fjz0             = _mm256_add_ps(fjz0,tz);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r30              = _mm256_mul_ps(rsq30,rinv30);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq30             = _mm256_mul_ps(iq3,jq0);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Analytical PME correction */
 911             zeta2            = _mm256_mul_ps(beta2,rsq30);
 912             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 913             pmecorrF         = avx256_pmecorrF_f(zeta2);
 914             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 915             felec            = _mm256_mul_ps(qq30,felec);
 916
 917             fscal            = felec;
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm256_mul_ps(fscal,dx30);
 921             ty               = _mm256_mul_ps(fscal,dy30);
 922             tz               = _mm256_mul_ps(fscal,dz30);
 923
 924             /* Update vectorial force */
 925             fix3             = _mm256_add_ps(fix3,tx);
 926             fiy3             = _mm256_add_ps(fiy3,ty);
 927             fiz3             = _mm256_add_ps(fiz3,tz);
 928
 929             fjx0             = _mm256_add_ps(fjx0,tx);
 930             fjy0             = _mm256_add_ps(fjy0,ty);
 931             fjz0             = _mm256_add_ps(fjz0,tz);
 932
 933             fjptrA             = f+j_coord_offsetA;
 934             fjptrB             = f+j_coord_offsetB;
 935             fjptrC             = f+j_coord_offsetC;
 936             fjptrD             = f+j_coord_offsetD;
 937             fjptrE             = f+j_coord_offsetE;
 938             fjptrF             = f+j_coord_offsetF;
 939             fjptrG             = f+j_coord_offsetG;
 940             fjptrH             = f+j_coord_offsetH;
 941
 942             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 943
 944             /* Inner loop uses 171 flops */
 945         }
 946
 947         if(jidx<j_index_end)
 948         {
 949
 950             /* Get j neighbor index, and coordinate index */
 951             jnrlistA         = jjnr[jidx];
 952             jnrlistB         = jjnr[jidx+1];
 953             jnrlistC         = jjnr[jidx+2];
 954             jnrlistD         = jjnr[jidx+3];
 955             jnrlistE         = jjnr[jidx+4];
 956             jnrlistF         = jjnr[jidx+5];
 957             jnrlistG         = jjnr[jidx+6];
 958             jnrlistH         = jjnr[jidx+7];
 959             /* Sign of each element will be negative for non-real atoms.
 960              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 961              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 962              */
 963             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 964                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 965
 966             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 967             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 968             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 969             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 970             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 971             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 972             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 973             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 974             j_coord_offsetA  = DIM*jnrA;
 975             j_coord_offsetB  = DIM*jnrB;
 976             j_coord_offsetC  = DIM*jnrC;
 977             j_coord_offsetD  = DIM*jnrD;
 978             j_coord_offsetE  = DIM*jnrE;
 979             j_coord_offsetF  = DIM*jnrF;
 980             j_coord_offsetG  = DIM*jnrG;
 981             j_coord_offsetH  = DIM*jnrH;
 982
 983             /* load j atom coordinates */
 984             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 985                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 986                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 987                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 988                                                  &jx0,&jy0,&jz0);
 989
 990             /* Calculate displacement vector */
 991             dx10             = _mm256_sub_ps(ix1,jx0);
 992             dy10             = _mm256_sub_ps(iy1,jy0);
 993             dz10             = _mm256_sub_ps(iz1,jz0);
 994             dx20             = _mm256_sub_ps(ix2,jx0);
 995             dy20             = _mm256_sub_ps(iy2,jy0);
 996             dz20             = _mm256_sub_ps(iz2,jz0);
 997             dx30             = _mm256_sub_ps(ix3,jx0);
 998             dy30             = _mm256_sub_ps(iy3,jy0);
 999             dz30             = _mm256_sub_ps(iz3,jz0);
1000
1001             /* Calculate squared distance and things based on it */
1002             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1003             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1004             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1005
1006             rinv10           = avx256_invsqrt_f(rsq10);
1007             rinv20           = avx256_invsqrt_f(rsq20);
1008             rinv30           = avx256_invsqrt_f(rsq30);
1009
1010             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1011             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1012             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1013
1014             /* Load parameters for j particles */
1015             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1016                                                                  charge+jnrC+0,charge+jnrD+0,
1017                                                                  charge+jnrE+0,charge+jnrF+0,
1018                                                                  charge+jnrG+0,charge+jnrH+0);
1019
1020             fjx0             = _mm256_setzero_ps();
1021             fjy0             = _mm256_setzero_ps();
1022             fjz0             = _mm256_setzero_ps();
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             r10              = _mm256_mul_ps(rsq10,rinv10);
1029             r10              = _mm256_andnot_ps(dummy_mask,r10);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm256_mul_ps(iq1,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Analytical PME correction */
1037             zeta2            = _mm256_mul_ps(beta2,rsq10);
1038             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1039             pmecorrF         = avx256_pmecorrF_f(zeta2);
1040             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1041             felec            = _mm256_mul_ps(qq10,felec);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm256_mul_ps(fscal,dx10);
1049             ty               = _mm256_mul_ps(fscal,dy10);
1050             tz               = _mm256_mul_ps(fscal,dz10);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm256_add_ps(fix1,tx);
1054             fiy1             = _mm256_add_ps(fiy1,ty);
1055             fiz1             = _mm256_add_ps(fiz1,tz);
1056
1057             fjx0             = _mm256_add_ps(fjx0,tx);
1058             fjy0             = _mm256_add_ps(fjy0,ty);
1059             fjz0             = _mm256_add_ps(fjz0,tz);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r20              = _mm256_mul_ps(rsq20,rinv20);
1066             r20              = _mm256_andnot_ps(dummy_mask,r20);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq20             = _mm256_mul_ps(iq2,jq0);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Analytical PME correction */
1074             zeta2            = _mm256_mul_ps(beta2,rsq20);
1075             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1076             pmecorrF         = avx256_pmecorrF_f(zeta2);
1077             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1078             felec            = _mm256_mul_ps(qq20,felec);
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_ps(fscal,dx20);
1086             ty               = _mm256_mul_ps(fscal,dy20);
1087             tz               = _mm256_mul_ps(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm256_add_ps(fix2,tx);
1091             fiy2             = _mm256_add_ps(fiy2,ty);
1092             fiz2             = _mm256_add_ps(fiz2,tz);
1093
1094             fjx0             = _mm256_add_ps(fjx0,tx);
1095             fjy0             = _mm256_add_ps(fjy0,ty);
1096             fjz0             = _mm256_add_ps(fjz0,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             r30              = _mm256_mul_ps(rsq30,rinv30);
1103             r30              = _mm256_andnot_ps(dummy_mask,r30);
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             qq30             = _mm256_mul_ps(iq3,jq0);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm256_mul_ps(beta2,rsq30);
1112             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1113             pmecorrF         = avx256_pmecorrF_f(zeta2);
1114             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1115             felec            = _mm256_mul_ps(qq30,felec);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_ps(fscal,dx30);
1123             ty               = _mm256_mul_ps(fscal,dy30);
1124             tz               = _mm256_mul_ps(fscal,dz30);
1125
1126             /* Update vectorial force */
1127             fix3             = _mm256_add_ps(fix3,tx);
1128             fiy3             = _mm256_add_ps(fiy3,ty);
1129             fiz3             = _mm256_add_ps(fiz3,tz);
1130
1131             fjx0             = _mm256_add_ps(fjx0,tx);
1132             fjy0             = _mm256_add_ps(fjy0,ty);
1133             fjz0             = _mm256_add_ps(fjz0,tz);
1134
1135             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1136             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1137             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1138             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1139             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1140             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1141             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1142             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1143
1144             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 174 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1152                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1153
1154         /* Increment number of inner iterations */
1155         inneriter                  += j_index_end - j_index_start;
1156
1157         /* Outer loop uses 18 flops */
1158     }
1159
1160     /* Increment number of outer iterations */
1161     outeriter        += nri;
1162
1163     /* Update outer/inner flops */
1164
1165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*174);
1166 }