src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     __m256i          ewitab;
 102     __m128i          ewitab_lo,ewitab_hi;
 103     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123
 124     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 125     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 126     beta2            = _mm256_mul_ps(beta,beta);
 127     beta3            = _mm256_mul_ps(beta,beta2);
 128
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 136     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 137     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->rcoulomb;
 141     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 142     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 179                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 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         fix3             = _mm256_setzero_ps();
 188         fiy3             = _mm256_setzero_ps();
 189         fiz3             = _mm256_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             jnrE             = jjnr[jidx+4];
 204             jnrF             = jjnr[jidx+5];
 205             jnrG             = jjnr[jidx+6];
 206             jnrH             = jjnr[jidx+7];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211             j_coord_offsetE  = DIM*jnrE;
 212             j_coord_offsetF  = DIM*jnrF;
 213             j_coord_offsetG  = DIM*jnrG;
 214             j_coord_offsetH  = DIM*jnrH;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 221                                                  &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 231             dy30             = _mm256_sub_ps(iy3,jy0);
 232             dz30             = _mm256_sub_ps(iz3,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 237             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 238
 239             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 240             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 241             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 242
 243             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 244             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 245             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                                  charge+jnrC+0,charge+jnrD+0,
 250                                                                  charge+jnrE+0,charge+jnrF+0,
 251                                                                  charge+jnrG+0,charge+jnrH+0);
 252
 253             fjx0             = _mm256_setzero_ps();
 254             fjy0             = _mm256_setzero_ps();
 255             fjz0             = _mm256_setzero_ps();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 262             {
 263
 264             r10              = _mm256_mul_ps(rsq10,rinv10);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq10             = _mm256_mul_ps(iq1,jq0);
 268
 269             /* EWALD ELECTROSTATICS */
 270
 271             /* Analytical PME correction */
 272             zeta2            = _mm256_mul_ps(beta2,rsq10);
 273             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 274             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 275             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 276             felec            = _mm256_mul_ps(qq10,felec);
 277             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 278             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 279             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 280             velec            = _mm256_mul_ps(qq10,velec);
 281
 282             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velec            = _mm256_and_ps(velec,cutoff_mask);
 286             velecsum         = _mm256_add_ps(velecsum,velec);
 287
 288             fscal            = felec;
 289
 290             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_ps(fscal,dx10);
 294             ty               = _mm256_mul_ps(fscal,dy10);
 295             tz               = _mm256_mul_ps(fscal,dz10);
 296
 297             /* Update vectorial force */
 298             fix1             = _mm256_add_ps(fix1,tx);
 299             fiy1             = _mm256_add_ps(fiy1,ty);
 300             fiz1             = _mm256_add_ps(fiz1,tz);
 301
 302             fjx0             = _mm256_add_ps(fjx0,tx);
 303             fjy0             = _mm256_add_ps(fjy0,ty);
 304             fjz0             = _mm256_add_ps(fjz0,tz);
 305
 306             }
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 313             {
 314
 315             r20              = _mm256_mul_ps(rsq20,rinv20);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm256_mul_ps(iq2,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Analytical PME correction */
 323             zeta2            = _mm256_mul_ps(beta2,rsq20);
 324             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 325             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 326             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 327             felec            = _mm256_mul_ps(qq20,felec);
 328             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 329             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 330             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 331             velec            = _mm256_mul_ps(qq20,velec);
 332
 333             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm256_and_ps(velec,cutoff_mask);
 337             velecsum         = _mm256_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm256_mul_ps(fscal,dx20);
 345             ty               = _mm256_mul_ps(fscal,dy20);
 346             tz               = _mm256_mul_ps(fscal,dz20);
 347
 348             /* Update vectorial force */
 349             fix2             = _mm256_add_ps(fix2,tx);
 350             fiy2             = _mm256_add_ps(fiy2,ty);
 351             fiz2             = _mm256_add_ps(fiz2,tz);
 352
 353             fjx0             = _mm256_add_ps(fjx0,tx);
 354             fjy0             = _mm256_add_ps(fjy0,ty);
 355             fjz0             = _mm256_add_ps(fjz0,tz);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 364             {
 365
 366             r30              = _mm256_mul_ps(rsq30,rinv30);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq30             = _mm256_mul_ps(iq3,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Analytical PME correction */
 374             zeta2            = _mm256_mul_ps(beta2,rsq30);
 375             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 376             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 377             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 378             felec            = _mm256_mul_ps(qq30,felec);
 379             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 380             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 381             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 382             velec            = _mm256_mul_ps(qq30,velec);
 383
 384             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm256_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm256_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_ps(fscal,dx30);
 396             ty               = _mm256_mul_ps(fscal,dy30);
 397             tz               = _mm256_mul_ps(fscal,dz30);
 398
 399             /* Update vectorial force */
 400             fix3             = _mm256_add_ps(fix3,tx);
 401             fiy3             = _mm256_add_ps(fiy3,ty);
 402             fiz3             = _mm256_add_ps(fiz3,tz);
 403
 404             fjx0             = _mm256_add_ps(fjx0,tx);
 405             fjy0             = _mm256_add_ps(fjy0,ty);
 406             fjz0             = _mm256_add_ps(fjz0,tz);
 407
 408             }
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414             fjptrE             = f+j_coord_offsetE;
 415             fjptrF             = f+j_coord_offsetF;
 416             fjptrG             = f+j_coord_offsetG;
 417             fjptrH             = f+j_coord_offsetH;
 418
 419             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 330 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             jnrlistE         = jjnr[jidx+4];
 433             jnrlistF         = jjnr[jidx+5];
 434             jnrlistG         = jjnr[jidx+6];
 435             jnrlistH         = jjnr[jidx+7];
 436             /* Sign of each element will be negative for non-real atoms.
 437              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 438              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 439              */
 440             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 441                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 442
 443             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 444             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 445             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 446             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 447             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 448             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 449             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 450             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 451             j_coord_offsetA  = DIM*jnrA;
 452             j_coord_offsetB  = DIM*jnrB;
 453             j_coord_offsetC  = DIM*jnrC;
 454             j_coord_offsetD  = DIM*jnrD;
 455             j_coord_offsetE  = DIM*jnrE;
 456             j_coord_offsetF  = DIM*jnrF;
 457             j_coord_offsetG  = DIM*jnrG;
 458             j_coord_offsetH  = DIM*jnrH;
 459
 460             /* load j atom coordinates */
 461             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 463                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 464                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 465                                                  &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 475             dy30             = _mm256_sub_ps(iy3,jy0);
 476             dz30             = _mm256_sub_ps(iz3,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 480             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 481             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 482
 483             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 484             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 485             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 486
 487             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 488             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 489             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 493                                                                  charge+jnrC+0,charge+jnrD+0,
 494                                                                  charge+jnrE+0,charge+jnrF+0,
 495                                                                  charge+jnrG+0,charge+jnrH+0);
 496
 497             fjx0             = _mm256_setzero_ps();
 498             fjy0             = _mm256_setzero_ps();
 499             fjz0             = _mm256_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 506             {
 507
 508             r10              = _mm256_mul_ps(rsq10,rinv10);
 509             r10              = _mm256_andnot_ps(dummy_mask,r10);
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq10             = _mm256_mul_ps(iq1,jq0);
 513
 514             /* EWALD ELECTROSTATICS */
 515
 516             /* Analytical PME correction */
 517             zeta2            = _mm256_mul_ps(beta2,rsq10);
 518             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 519             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 520             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 521             felec            = _mm256_mul_ps(qq10,felec);
 522             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 523             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 524             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 525             velec            = _mm256_mul_ps(qq10,velec);
 526
 527             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm256_and_ps(velec,cutoff_mask);
 531             velec            = _mm256_andnot_ps(dummy_mask,velec);
 532             velecsum         = _mm256_add_ps(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 537
 538             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm256_mul_ps(fscal,dx10);
 542             ty               = _mm256_mul_ps(fscal,dy10);
 543             tz               = _mm256_mul_ps(fscal,dz10);
 544
 545             /* Update vectorial force */
 546             fix1             = _mm256_add_ps(fix1,tx);
 547             fiy1             = _mm256_add_ps(fiy1,ty);
 548             fiz1             = _mm256_add_ps(fiz1,tz);
 549
 550             fjx0             = _mm256_add_ps(fjx0,tx);
 551             fjy0             = _mm256_add_ps(fjy0,ty);
 552             fjz0             = _mm256_add_ps(fjz0,tz);
 553
 554             }
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 561             {
 562
 563             r20              = _mm256_mul_ps(rsq20,rinv20);
 564             r20              = _mm256_andnot_ps(dummy_mask,r20);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq20             = _mm256_mul_ps(iq2,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Analytical PME correction */
 572             zeta2            = _mm256_mul_ps(beta2,rsq20);
 573             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 574             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 575             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 576             felec            = _mm256_mul_ps(qq20,felec);
 577             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 578             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 579             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 580             velec            = _mm256_mul_ps(qq20,velec);
 581
 582             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm256_and_ps(velec,cutoff_mask);
 586             velec            = _mm256_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx20);
 597             ty               = _mm256_mul_ps(fscal,dy20);
 598             tz               = _mm256_mul_ps(fscal,dz20);
 599
 600             /* Update vectorial force */
 601             fix2             = _mm256_add_ps(fix2,tx);
 602             fiy2             = _mm256_add_ps(fiy2,ty);
 603             fiz2             = _mm256_add_ps(fiz2,tz);
 604
 605             fjx0             = _mm256_add_ps(fjx0,tx);
 606             fjy0             = _mm256_add_ps(fjy0,ty);
 607             fjz0             = _mm256_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 616             {
 617
 618             r30              = _mm256_mul_ps(rsq30,rinv30);
 619             r30              = _mm256_andnot_ps(dummy_mask,r30);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq30             = _mm256_mul_ps(iq3,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Analytical PME correction */
 627             zeta2            = _mm256_mul_ps(beta2,rsq30);
 628             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 629             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 630             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 631             felec            = _mm256_mul_ps(qq30,felec);
 632             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 633             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 634             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 635             velec            = _mm256_mul_ps(qq30,velec);
 636
 637             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_and_ps(velec,cutoff_mask);
 641             velec            = _mm256_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 647
 648             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_ps(fscal,dx30);
 652             ty               = _mm256_mul_ps(fscal,dy30);
 653             tz               = _mm256_mul_ps(fscal,dz30);
 654
 655             /* Update vectorial force */
 656             fix3             = _mm256_add_ps(fix3,tx);
 657             fiy3             = _mm256_add_ps(fiy3,ty);
 658             fiz3             = _mm256_add_ps(fiz3,tz);
 659
 660             fjx0             = _mm256_add_ps(fjx0,tx);
 661             fjy0             = _mm256_add_ps(fjy0,ty);
 662             fjz0             = _mm256_add_ps(fjz0,tz);
 663
 664             }
 665
 666             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 667             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 668             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 669             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 670             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 671             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 672             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 673             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 674
 675             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 676
 677             /* Inner loop uses 333 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 683                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 684
 685         ggid                        = gid[iidx];
 686         /* Update potential energies */
 687         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 688
 689         /* Increment number of inner iterations */
 690         inneriter                  += j_index_end - j_index_start;
 691
 692         /* Outer loop uses 19 flops */
 693     }
 694
 695     /* Increment number of outer iterations */
 696     outeriter        += nri;
 697
 698     /* Update outer/inner flops */
 699
 700     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*333);
 701 }
 702 /*
 703  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_256_single
 704  * Electrostatics interaction: Ewald
 705  * VdW interaction:            None
 706  * Geometry:                   Water4-Particle
 707  * Calculate force/pot:        Force
 708  */
 709 void
 710 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_256_single
 711                     (t_nblist                    * gmx_restrict       nlist,
 712                      rvec                        * gmx_restrict          xx,
 713                      rvec                        * gmx_restrict          ff,
 714                      t_forcerec                  * gmx_restrict          fr,
 715                      t_mdatoms                   * gmx_restrict     mdatoms,
 716                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 717                      t_nrnb                      * gmx_restrict        nrnb)
 718 {
 719     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 720      * just 0 for non-waters.
 721      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 722      * jnr indices corresponding to data put in the four positions in the SIMD register.
 723      */
 724     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 725     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 726     int              jnrA,jnrB,jnrC,jnrD;
 727     int              jnrE,jnrF,jnrG,jnrH;
 728     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 729     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 730     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 731     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 732     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 733     real             rcutoff_scalar;
 734     real             *shiftvec,*fshift,*x,*f;
 735     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 736     real             scratch[4*DIM];
 737     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 738     real *           vdwioffsetptr1;
 739     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     real *           vdwioffsetptr2;
 741     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     real *           vdwioffsetptr3;
 743     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 744     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 745     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 746     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 747     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 748     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 749     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 750     real             *charge;
 751     __m256i          ewitab;
 752     __m128i          ewitab_lo,ewitab_hi;
 753     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 754     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 755     real             *ewtab;
 756     __m256           dummy_mask,cutoff_mask;
 757     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 758     __m256           one     = _mm256_set1_ps(1.0);
 759     __m256           two     = _mm256_set1_ps(2.0);
 760     x                = xx[0];
 761     f                = ff[0];
 762
 763     nri              = nlist->nri;
 764     iinr             = nlist->iinr;
 765     jindex           = nlist->jindex;
 766     jjnr             = nlist->jjnr;
 767     shiftidx         = nlist->shift;
 768     gid              = nlist->gid;
 769     shiftvec         = fr->shift_vec[0];
 770     fshift           = fr->fshift[0];
 771     facel            = _mm256_set1_ps(fr->epsfac);
 772     charge           = mdatoms->chargeA;
 773
 774     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 775     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 776     beta2            = _mm256_mul_ps(beta,beta);
 777     beta3            = _mm256_mul_ps(beta,beta2);
 778
 779     ewtab            = fr->ic->tabq_coul_F;
 780     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 781     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 782
 783     /* Setup water-specific parameters */
 784     inr              = nlist->iinr[0];
 785     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 786     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 787     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 788
 789     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 790     rcutoff_scalar   = fr->rcoulomb;
 791     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 792     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 793
 794     /* Avoid stupid compiler warnings */
 795     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 796     j_coord_offsetA = 0;
 797     j_coord_offsetB = 0;
 798     j_coord_offsetC = 0;
 799     j_coord_offsetD = 0;
 800     j_coord_offsetE = 0;
 801     j_coord_offsetF = 0;
 802     j_coord_offsetG = 0;
 803     j_coord_offsetH = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     for(iidx=0;iidx<4*DIM;iidx++)
 809     {
 810         scratch[iidx] = 0.0;
 811     }
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 829                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 830
 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         fix3             = _mm256_setzero_ps();
 838         fiy3             = _mm256_setzero_ps();
 839         fiz3             = _mm256_setzero_ps();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             jnrC             = jjnr[jidx+2];
 849             jnrD             = jjnr[jidx+3];
 850             jnrE             = jjnr[jidx+4];
 851             jnrF             = jjnr[jidx+5];
 852             jnrG             = jjnr[jidx+6];
 853             jnrH             = jjnr[jidx+7];
 854             j_coord_offsetA  = DIM*jnrA;
 855             j_coord_offsetB  = DIM*jnrB;
 856             j_coord_offsetC  = DIM*jnrC;
 857             j_coord_offsetD  = DIM*jnrD;
 858             j_coord_offsetE  = DIM*jnrE;
 859             j_coord_offsetF  = DIM*jnrF;
 860             j_coord_offsetG  = DIM*jnrG;
 861             j_coord_offsetH  = DIM*jnrH;
 862
 863             /* load j atom coordinates */
 864             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 865                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 866                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 867                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 868                                                  &jx0,&jy0,&jz0);
 869
 870             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 878             dy30             = _mm256_sub_ps(iy3,jy0);
 879             dz30             = _mm256_sub_ps(iz3,jz0);
 880
 881             /* Calculate squared distance and things based on it */
 882             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 883             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 884             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 885
 886             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 887             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 888             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 889
 890             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 891             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 892             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 893
 894             /* Load parameters for j particles */
 895             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 896                                                                  charge+jnrC+0,charge+jnrD+0,
 897                                                                  charge+jnrE+0,charge+jnrF+0,
 898                                                                  charge+jnrG+0,charge+jnrH+0);
 899
 900             fjx0             = _mm256_setzero_ps();
 901             fjy0             = _mm256_setzero_ps();
 902             fjz0             = _mm256_setzero_ps();
 903
 904             /**************************
 905              * CALCULATE INTERACTIONS *
 906              **************************/
 907
 908             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 909             {
 910
 911             r10              = _mm256_mul_ps(rsq10,rinv10);
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq10             = _mm256_mul_ps(iq1,jq0);
 915
 916             /* EWALD ELECTROSTATICS */
 917
 918             /* Analytical PME correction */
 919             zeta2            = _mm256_mul_ps(beta2,rsq10);
 920             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 921             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 922             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 923             felec            = _mm256_mul_ps(qq10,felec);
 924
 925             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 926
 927             fscal            = felec;
 928
 929             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm256_mul_ps(fscal,dx10);
 933             ty               = _mm256_mul_ps(fscal,dy10);
 934             tz               = _mm256_mul_ps(fscal,dz10);
 935
 936             /* Update vectorial force */
 937             fix1             = _mm256_add_ps(fix1,tx);
 938             fiy1             = _mm256_add_ps(fiy1,ty);
 939             fiz1             = _mm256_add_ps(fiz1,tz);
 940
 941             fjx0             = _mm256_add_ps(fjx0,tx);
 942             fjy0             = _mm256_add_ps(fjy0,ty);
 943             fjz0             = _mm256_add_ps(fjz0,tz);
 944
 945             }
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 952             {
 953
 954             r20              = _mm256_mul_ps(rsq20,rinv20);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq20             = _mm256_mul_ps(iq2,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Analytical PME correction */
 962             zeta2            = _mm256_mul_ps(beta2,rsq20);
 963             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 964             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 965             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 966             felec            = _mm256_mul_ps(qq20,felec);
 967
 968             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_ps(fscal,dx20);
 976             ty               = _mm256_mul_ps(fscal,dy20);
 977             tz               = _mm256_mul_ps(fscal,dz20);
 978
 979             /* Update vectorial force */
 980             fix2             = _mm256_add_ps(fix2,tx);
 981             fiy2             = _mm256_add_ps(fiy2,ty);
 982             fiz2             = _mm256_add_ps(fiz2,tz);
 983
 984             fjx0             = _mm256_add_ps(fjx0,tx);
 985             fjy0             = _mm256_add_ps(fjy0,ty);
 986             fjz0             = _mm256_add_ps(fjz0,tz);
 987
 988             }
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 995             {
 996
 997             r30              = _mm256_mul_ps(rsq30,rinv30);
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq30             = _mm256_mul_ps(iq3,jq0);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Analytical PME correction */
1005             zeta2            = _mm256_mul_ps(beta2,rsq30);
1006             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1007             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1008             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1009             felec            = _mm256_mul_ps(qq30,felec);
1010
1011             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1012
1013             fscal            = felec;
1014
1015             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm256_mul_ps(fscal,dx30);
1019             ty               = _mm256_mul_ps(fscal,dy30);
1020             tz               = _mm256_mul_ps(fscal,dz30);
1021
1022             /* Update vectorial force */
1023             fix3             = _mm256_add_ps(fix3,tx);
1024             fiy3             = _mm256_add_ps(fiy3,ty);
1025             fiz3             = _mm256_add_ps(fiz3,tz);
1026
1027             fjx0             = _mm256_add_ps(fjx0,tx);
1028             fjy0             = _mm256_add_ps(fjy0,ty);
1029             fjz0             = _mm256_add_ps(fjz0,tz);
1030
1031             }
1032
1033             fjptrA             = f+j_coord_offsetA;
1034             fjptrB             = f+j_coord_offsetB;
1035             fjptrC             = f+j_coord_offsetC;
1036             fjptrD             = f+j_coord_offsetD;
1037             fjptrE             = f+j_coord_offsetE;
1038             fjptrF             = f+j_coord_offsetF;
1039             fjptrG             = f+j_coord_offsetG;
1040             fjptrH             = f+j_coord_offsetH;
1041
1042             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1043
1044             /* Inner loop uses 180 flops */
1045         }
1046
1047         if(jidx<j_index_end)
1048         {
1049
1050             /* Get j neighbor index, and coordinate index */
1051             jnrlistA         = jjnr[jidx];
1052             jnrlistB         = jjnr[jidx+1];
1053             jnrlistC         = jjnr[jidx+2];
1054             jnrlistD         = jjnr[jidx+3];
1055             jnrlistE         = jjnr[jidx+4];
1056             jnrlistF         = jjnr[jidx+5];
1057             jnrlistG         = jjnr[jidx+6];
1058             jnrlistH         = jjnr[jidx+7];
1059             /* Sign of each element will be negative for non-real atoms.
1060              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1061              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1062              */
1063             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1064                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1065
1066             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1067             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1068             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1069             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1070             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1071             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1072             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1073             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1074             j_coord_offsetA  = DIM*jnrA;
1075             j_coord_offsetB  = DIM*jnrB;
1076             j_coord_offsetC  = DIM*jnrC;
1077             j_coord_offsetD  = DIM*jnrD;
1078             j_coord_offsetE  = DIM*jnrE;
1079             j_coord_offsetF  = DIM*jnrF;
1080             j_coord_offsetG  = DIM*jnrG;
1081             j_coord_offsetH  = DIM*jnrH;
1082
1083             /* load j atom coordinates */
1084             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1085                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1086                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1087                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1088                                                  &jx0,&jy0,&jz0);
1089
1090             /* Calculate displacement vector */
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             dx30             = _mm256_sub_ps(ix3,jx0);
1098             dy30             = _mm256_sub_ps(iy3,jy0);
1099             dz30             = _mm256_sub_ps(iz3,jz0);
1100
1101             /* Calculate squared distance and things based on it */
1102             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1103             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1104             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1105
1106             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1107             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1108             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1109
1110             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1111             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1112             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1113
1114             /* Load parameters for j particles */
1115             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1116                                                                  charge+jnrC+0,charge+jnrD+0,
1117                                                                  charge+jnrE+0,charge+jnrF+0,
1118                                                                  charge+jnrG+0,charge+jnrH+0);
1119
1120             fjx0             = _mm256_setzero_ps();
1121             fjy0             = _mm256_setzero_ps();
1122             fjz0             = _mm256_setzero_ps();
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1129             {
1130
1131             r10              = _mm256_mul_ps(rsq10,rinv10);
1132             r10              = _mm256_andnot_ps(dummy_mask,r10);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq10             = _mm256_mul_ps(iq1,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Analytical PME correction */
1140             zeta2            = _mm256_mul_ps(beta2,rsq10);
1141             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1142             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1143             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1144             felec            = _mm256_mul_ps(qq10,felec);
1145
1146             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1151
1152             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_ps(fscal,dx10);
1156             ty               = _mm256_mul_ps(fscal,dy10);
1157             tz               = _mm256_mul_ps(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm256_add_ps(fix1,tx);
1161             fiy1             = _mm256_add_ps(fiy1,ty);
1162             fiz1             = _mm256_add_ps(fiz1,tz);
1163
1164             fjx0             = _mm256_add_ps(fjx0,tx);
1165             fjy0             = _mm256_add_ps(fjy0,ty);
1166             fjz0             = _mm256_add_ps(fjz0,tz);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1175             {
1176
1177             r20              = _mm256_mul_ps(rsq20,rinv20);
1178             r20              = _mm256_andnot_ps(dummy_mask,r20);
1179
1180             /* Compute parameters for interactions between i and j atoms */
1181             qq20             = _mm256_mul_ps(iq2,jq0);
1182
1183             /* EWALD ELECTROSTATICS */
1184
1185             /* Analytical PME correction */
1186             zeta2            = _mm256_mul_ps(beta2,rsq20);
1187             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1188             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1189             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1190             felec            = _mm256_mul_ps(qq20,felec);
1191
1192             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1193
1194             fscal            = felec;
1195
1196             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1197
1198             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm256_mul_ps(fscal,dx20);
1202             ty               = _mm256_mul_ps(fscal,dy20);
1203             tz               = _mm256_mul_ps(fscal,dz20);
1204
1205             /* Update vectorial force */
1206             fix2             = _mm256_add_ps(fix2,tx);
1207             fiy2             = _mm256_add_ps(fiy2,ty);
1208             fiz2             = _mm256_add_ps(fiz2,tz);
1209
1210             fjx0             = _mm256_add_ps(fjx0,tx);
1211             fjy0             = _mm256_add_ps(fjy0,ty);
1212             fjz0             = _mm256_add_ps(fjz0,tz);
1213
1214             }
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1221             {
1222
1223             r30              = _mm256_mul_ps(rsq30,rinv30);
1224             r30              = _mm256_andnot_ps(dummy_mask,r30);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm256_mul_ps(iq3,jq0);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Analytical PME correction */
1232             zeta2            = _mm256_mul_ps(beta2,rsq30);
1233             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1234             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1235             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1236             felec            = _mm256_mul_ps(qq30,felec);
1237
1238             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1243
1244             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm256_mul_ps(fscal,dx30);
1248             ty               = _mm256_mul_ps(fscal,dy30);
1249             tz               = _mm256_mul_ps(fscal,dz30);
1250
1251             /* Update vectorial force */
1252             fix3             = _mm256_add_ps(fix3,tx);
1253             fiy3             = _mm256_add_ps(fiy3,ty);
1254             fiz3             = _mm256_add_ps(fiz3,tz);
1255
1256             fjx0             = _mm256_add_ps(fjx0,tx);
1257             fjy0             = _mm256_add_ps(fjy0,ty);
1258             fjz0             = _mm256_add_ps(fjz0,tz);
1259
1260             }
1261
1262             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1263             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1264             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1265             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1266             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1267             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1268             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1269             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1270
1271             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1272
1273             /* Inner loop uses 183 flops */
1274         }
1275
1276         /* End of innermost loop */
1277
1278         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1279                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1280
1281         /* Increment number of inner iterations */
1282         inneriter                  += j_index_end - j_index_start;
1283
1284         /* Outer loop uses 18 flops */
1285     }
1286
1287     /* Increment number of outer iterations */
1288     outeriter        += nri;
1289
1290     /* Update outer/inner flops */
1291
1292     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*183);
1293 }