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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 123     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 124     beta2            = _mm256_mul_ps(beta,beta);
 125     beta3            = _mm256_mul_ps(beta,beta2);
 126
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 134     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 135     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148     j_coord_offsetE = 0;
 149     j_coord_offsetF = 0;
 150     j_coord_offsetG = 0;
 151     j_coord_offsetH = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 178
 179         fix0             = _mm256_setzero_ps();
 180         fiy0             = _mm256_setzero_ps();
 181         fiz0             = _mm256_setzero_ps();
 182         fix1             = _mm256_setzero_ps();
 183         fiy1             = _mm256_setzero_ps();
 184         fiz1             = _mm256_setzero_ps();
 185         fix2             = _mm256_setzero_ps();
 186         fiy2             = _mm256_setzero_ps();
 187         fiz2             = _mm256_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250
 251             fjx0             = _mm256_setzero_ps();
 252             fjy0             = _mm256_setzero_ps();
 253             fjz0             = _mm256_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 260             {
 261
 262             r00              = _mm256_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm256_mul_ps(iq0,jq0);
 266
 267             /* EWALD ELECTROSTATICS */
 268
 269             /* Analytical PME correction */
 270             zeta2            = _mm256_mul_ps(beta2,rsq00);
 271             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 272             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 273             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 274             felec            = _mm256_mul_ps(qq00,felec);
 275             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 276             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 277             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 278             velec            = _mm256_mul_ps(qq00,velec);
 279
 280             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velec            = _mm256_and_ps(velec,cutoff_mask);
 284             velecsum         = _mm256_add_ps(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 289
 290             /* Calculate temporary vectorial force */
 291             tx               = _mm256_mul_ps(fscal,dx00);
 292             ty               = _mm256_mul_ps(fscal,dy00);
 293             tz               = _mm256_mul_ps(fscal,dz00);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm256_add_ps(fix0,tx);
 297             fiy0             = _mm256_add_ps(fiy0,ty);
 298             fiz0             = _mm256_add_ps(fiz0,tz);
 299
 300             fjx0             = _mm256_add_ps(fjx0,tx);
 301             fjy0             = _mm256_add_ps(fjy0,ty);
 302             fjz0             = _mm256_add_ps(fjz0,tz);
 303
 304             }
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 311             {
 312
 313             r10              = _mm256_mul_ps(rsq10,rinv10);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm256_mul_ps(iq1,jq0);
 317
 318             /* EWALD ELECTROSTATICS */
 319
 320             /* Analytical PME correction */
 321             zeta2            = _mm256_mul_ps(beta2,rsq10);
 322             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 323             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 324             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 325             felec            = _mm256_mul_ps(qq10,felec);
 326             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 327             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 328             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 329             velec            = _mm256_mul_ps(qq10,velec);
 330
 331             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm256_and_ps(velec,cutoff_mask);
 335             velecsum         = _mm256_add_ps(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm256_mul_ps(fscal,dx10);
 343             ty               = _mm256_mul_ps(fscal,dy10);
 344             tz               = _mm256_mul_ps(fscal,dz10);
 345
 346             /* Update vectorial force */
 347             fix1             = _mm256_add_ps(fix1,tx);
 348             fiy1             = _mm256_add_ps(fiy1,ty);
 349             fiz1             = _mm256_add_ps(fiz1,tz);
 350
 351             fjx0             = _mm256_add_ps(fjx0,tx);
 352             fjy0             = _mm256_add_ps(fjy0,ty);
 353             fjz0             = _mm256_add_ps(fjz0,tz);
 354
 355             }
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 362             {
 363
 364             r20              = _mm256_mul_ps(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm256_mul_ps(iq2,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Analytical PME correction */
 372             zeta2            = _mm256_mul_ps(beta2,rsq20);
 373             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 374             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 375             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 376             felec            = _mm256_mul_ps(qq20,felec);
 377             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 378             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 379             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 380             velec            = _mm256_mul_ps(qq20,velec);
 381
 382             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm256_and_ps(velec,cutoff_mask);
 386             velecsum         = _mm256_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm256_mul_ps(fscal,dx20);
 394             ty               = _mm256_mul_ps(fscal,dy20);
 395             tz               = _mm256_mul_ps(fscal,dz20);
 396
 397             /* Update vectorial force */
 398             fix2             = _mm256_add_ps(fix2,tx);
 399             fiy2             = _mm256_add_ps(fiy2,ty);
 400             fiz2             = _mm256_add_ps(fiz2,tz);
 401
 402             fjx0             = _mm256_add_ps(fjx0,tx);
 403             fjy0             = _mm256_add_ps(fjy0,ty);
 404             fjz0             = _mm256_add_ps(fjz0,tz);
 405
 406             }
 407
 408             fjptrA             = f+j_coord_offsetA;
 409             fjptrB             = f+j_coord_offsetB;
 410             fjptrC             = f+j_coord_offsetC;
 411             fjptrD             = f+j_coord_offsetD;
 412             fjptrE             = f+j_coord_offsetE;
 413             fjptrF             = f+j_coord_offsetF;
 414             fjptrG             = f+j_coord_offsetG;
 415             fjptrH             = f+j_coord_offsetH;
 416
 417             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 330 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             jnrlistE         = jjnr[jidx+4];
 431             jnrlistF         = jjnr[jidx+5];
 432             jnrlistG         = jjnr[jidx+6];
 433             jnrlistH         = jjnr[jidx+7];
 434             /* Sign of each element will be negative for non-real atoms.
 435              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 436              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 437              */
 438             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 439                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 440
 441             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 442             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 443             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 444             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 445             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 446             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 447             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 448             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 449             j_coord_offsetA  = DIM*jnrA;
 450             j_coord_offsetB  = DIM*jnrB;
 451             j_coord_offsetC  = DIM*jnrC;
 452             j_coord_offsetD  = DIM*jnrD;
 453             j_coord_offsetE  = DIM*jnrE;
 454             j_coord_offsetF  = DIM*jnrF;
 455             j_coord_offsetG  = DIM*jnrG;
 456             j_coord_offsetH  = DIM*jnrH;
 457
 458             /* load j atom coordinates */
 459             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 461                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 462                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 463                                                  &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm256_sub_ps(ix0,jx0);
 467             dy00             = _mm256_sub_ps(iy0,jy0);
 468             dz00             = _mm256_sub_ps(iz0,jz0);
 469             dx10             = _mm256_sub_ps(ix1,jx0);
 470             dy10             = _mm256_sub_ps(iy1,jy0);
 471             dz10             = _mm256_sub_ps(iz1,jz0);
 472             dx20             = _mm256_sub_ps(ix2,jx0);
 473             dy20             = _mm256_sub_ps(iy2,jy0);
 474             dz20             = _mm256_sub_ps(iz2,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 478             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 479             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 480
 481             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 482             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 483             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 484
 485             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 486             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 487             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                                  charge+jnrC+0,charge+jnrD+0,
 492                                                                  charge+jnrE+0,charge+jnrF+0,
 493                                                                  charge+jnrG+0,charge+jnrH+0);
 494
 495             fjx0             = _mm256_setzero_ps();
 496             fjy0             = _mm256_setzero_ps();
 497             fjz0             = _mm256_setzero_ps();
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 504             {
 505
 506             r00              = _mm256_mul_ps(rsq00,rinv00);
 507             r00              = _mm256_andnot_ps(dummy_mask,r00);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq00             = _mm256_mul_ps(iq0,jq0);
 511
 512             /* EWALD ELECTROSTATICS */
 513
 514             /* Analytical PME correction */
 515             zeta2            = _mm256_mul_ps(beta2,rsq00);
 516             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 517             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 518             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 519             felec            = _mm256_mul_ps(qq00,felec);
 520             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 521             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 522             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 523             velec            = _mm256_mul_ps(qq00,velec);
 524
 525             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm256_and_ps(velec,cutoff_mask);
 529             velec            = _mm256_andnot_ps(dummy_mask,velec);
 530             velecsum         = _mm256_add_ps(velecsum,velec);
 531
 532             fscal            = felec;
 533
 534             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 535
 536             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_ps(fscal,dx00);
 540             ty               = _mm256_mul_ps(fscal,dy00);
 541             tz               = _mm256_mul_ps(fscal,dz00);
 542
 543             /* Update vectorial force */
 544             fix0             = _mm256_add_ps(fix0,tx);
 545             fiy0             = _mm256_add_ps(fiy0,ty);
 546             fiz0             = _mm256_add_ps(fiz0,tz);
 547
 548             fjx0             = _mm256_add_ps(fjx0,tx);
 549             fjy0             = _mm256_add_ps(fjy0,ty);
 550             fjz0             = _mm256_add_ps(fjz0,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 559             {
 560
 561             r10              = _mm256_mul_ps(rsq10,rinv10);
 562             r10              = _mm256_andnot_ps(dummy_mask,r10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm256_mul_ps(iq1,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Analytical PME correction */
 570             zeta2            = _mm256_mul_ps(beta2,rsq10);
 571             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 572             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 573             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 574             felec            = _mm256_mul_ps(qq10,felec);
 575             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 576             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 577             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 578             velec            = _mm256_mul_ps(qq10,velec);
 579
 580             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_and_ps(velec,cutoff_mask);
 584             velec            = _mm256_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm256_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_ps(fscal,dx10);
 595             ty               = _mm256_mul_ps(fscal,dy10);
 596             tz               = _mm256_mul_ps(fscal,dz10);
 597
 598             /* Update vectorial force */
 599             fix1             = _mm256_add_ps(fix1,tx);
 600             fiy1             = _mm256_add_ps(fiy1,ty);
 601             fiz1             = _mm256_add_ps(fiz1,tz);
 602
 603             fjx0             = _mm256_add_ps(fjx0,tx);
 604             fjy0             = _mm256_add_ps(fjy0,ty);
 605             fjz0             = _mm256_add_ps(fjz0,tz);
 606
 607             }
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 614             {
 615
 616             r20              = _mm256_mul_ps(rsq20,rinv20);
 617             r20              = _mm256_andnot_ps(dummy_mask,r20);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq20             = _mm256_mul_ps(iq2,jq0);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Analytical PME correction */
 625             zeta2            = _mm256_mul_ps(beta2,rsq20);
 626             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 627             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 628             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 629             felec            = _mm256_mul_ps(qq20,felec);
 630             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 631             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 632             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 633             velec            = _mm256_mul_ps(qq20,velec);
 634
 635             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm256_and_ps(velec,cutoff_mask);
 639             velec            = _mm256_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm256_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 645
 646             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_ps(fscal,dx20);
 650             ty               = _mm256_mul_ps(fscal,dy20);
 651             tz               = _mm256_mul_ps(fscal,dz20);
 652
 653             /* Update vectorial force */
 654             fix2             = _mm256_add_ps(fix2,tx);
 655             fiy2             = _mm256_add_ps(fiy2,ty);
 656             fiz2             = _mm256_add_ps(fiz2,tz);
 657
 658             fjx0             = _mm256_add_ps(fjx0,tx);
 659             fjy0             = _mm256_add_ps(fjy0,ty);
 660             fjz0             = _mm256_add_ps(fjz0,tz);
 661
 662             }
 663
 664             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 665             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 666             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 667             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 668             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 669             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 670             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 671             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 672
 673             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 674
 675             /* Inner loop uses 333 flops */
 676         }
 677
 678         /* End of innermost loop */
 679
 680         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 681                                                  f+i_coord_offset,fshift+i_shift_offset);
 682
 683         ggid                        = gid[iidx];
 684         /* Update potential energies */
 685         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 19 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*333);
 699 }
 700 /*
 701  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 702  * Electrostatics interaction: Ewald
 703  * VdW interaction:            None
 704  * Geometry:                   Water3-Particle
 705  * Calculate force/pot:        Force
 706  */
 707 void
 708 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 709                     (t_nblist                    * gmx_restrict       nlist,
 710                      rvec                        * gmx_restrict          xx,
 711                      rvec                        * gmx_restrict          ff,
 712                      t_forcerec                  * gmx_restrict          fr,
 713                      t_mdatoms                   * gmx_restrict     mdatoms,
 714                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 715                      t_nrnb                      * gmx_restrict        nrnb)
 716 {
 717     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 718      * just 0 for non-waters.
 719      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 720      * jnr indices corresponding to data put in the four positions in the SIMD register.
 721      */
 722     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 723     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 724     int              jnrA,jnrB,jnrC,jnrD;
 725     int              jnrE,jnrF,jnrG,jnrH;
 726     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 727     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 728     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 729     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 734     real             scratch[4*DIM];
 735     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     real *           vdwioffsetptr0;
 737     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 743     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     __m256i          ewitab;
 750     __m128i          ewitab_lo,ewitab_hi;
 751     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 752     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 753     real             *ewtab;
 754     __m256           dummy_mask,cutoff_mask;
 755     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 756     __m256           one     = _mm256_set1_ps(1.0);
 757     __m256           two     = _mm256_set1_ps(2.0);
 758     x                = xx[0];
 759     f                = ff[0];
 760
 761     nri              = nlist->nri;
 762     iinr             = nlist->iinr;
 763     jindex           = nlist->jindex;
 764     jjnr             = nlist->jjnr;
 765     shiftidx         = nlist->shift;
 766     gid              = nlist->gid;
 767     shiftvec         = fr->shift_vec[0];
 768     fshift           = fr->fshift[0];
 769     facel            = _mm256_set1_ps(fr->epsfac);
 770     charge           = mdatoms->chargeA;
 771
 772     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 773     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 774     beta2            = _mm256_mul_ps(beta,beta);
 775     beta3            = _mm256_mul_ps(beta,beta2);
 776
 777     ewtab            = fr->ic->tabq_coul_F;
 778     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 779     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 780
 781     /* Setup water-specific parameters */
 782     inr              = nlist->iinr[0];
 783     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 784     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 785     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 786
 787     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 788     rcutoff_scalar   = fr->rcoulomb;
 789     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 790     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 791
 792     /* Avoid stupid compiler warnings */
 793     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 794     j_coord_offsetA = 0;
 795     j_coord_offsetB = 0;
 796     j_coord_offsetC = 0;
 797     j_coord_offsetD = 0;
 798     j_coord_offsetE = 0;
 799     j_coord_offsetF = 0;
 800     j_coord_offsetG = 0;
 801     j_coord_offsetH = 0;
 802
 803     outeriter        = 0;
 804     inneriter        = 0;
 805
 806     for(iidx=0;iidx<4*DIM;iidx++)
 807     {
 808         scratch[iidx] = 0.0;
 809     }
 810
 811     /* Start outer loop over neighborlists */
 812     for(iidx=0; iidx<nri; iidx++)
 813     {
 814         /* Load shift vector for this list */
 815         i_shift_offset   = DIM*shiftidx[iidx];
 816
 817         /* Load limits for loop over neighbors */
 818         j_index_start    = jindex[iidx];
 819         j_index_end      = jindex[iidx+1];
 820
 821         /* Get outer coordinate index */
 822         inr              = iinr[iidx];
 823         i_coord_offset   = DIM*inr;
 824
 825         /* Load i particle coords and add shift vector */
 826         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 827                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 828
 829         fix0             = _mm256_setzero_ps();
 830         fiy0             = _mm256_setzero_ps();
 831         fiz0             = _mm256_setzero_ps();
 832         fix1             = _mm256_setzero_ps();
 833         fiy1             = _mm256_setzero_ps();
 834         fiz1             = _mm256_setzero_ps();
 835         fix2             = _mm256_setzero_ps();
 836         fiy2             = _mm256_setzero_ps();
 837         fiz2             = _mm256_setzero_ps();
 838
 839         /* Start inner kernel loop */
 840         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrA             = jjnr[jidx];
 845             jnrB             = jjnr[jidx+1];
 846             jnrC             = jjnr[jidx+2];
 847             jnrD             = jjnr[jidx+3];
 848             jnrE             = jjnr[jidx+4];
 849             jnrF             = jjnr[jidx+5];
 850             jnrG             = jjnr[jidx+6];
 851             jnrH             = jjnr[jidx+7];
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854             j_coord_offsetC  = DIM*jnrC;
 855             j_coord_offsetD  = DIM*jnrD;
 856             j_coord_offsetE  = DIM*jnrE;
 857             j_coord_offsetF  = DIM*jnrF;
 858             j_coord_offsetG  = DIM*jnrG;
 859             j_coord_offsetH  = DIM*jnrH;
 860
 861             /* load j atom coordinates */
 862             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 863                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 864                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 865                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 866                                                  &jx0,&jy0,&jz0);
 867
 868             /* Calculate displacement vector */
 869             dx00             = _mm256_sub_ps(ix0,jx0);
 870             dy00             = _mm256_sub_ps(iy0,jy0);
 871             dz00             = _mm256_sub_ps(iz0,jz0);
 872             dx10             = _mm256_sub_ps(ix1,jx0);
 873             dy10             = _mm256_sub_ps(iy1,jy0);
 874             dz10             = _mm256_sub_ps(iz1,jz0);
 875             dx20             = _mm256_sub_ps(ix2,jx0);
 876             dy20             = _mm256_sub_ps(iy2,jy0);
 877             dz20             = _mm256_sub_ps(iz2,jz0);
 878
 879             /* Calculate squared distance and things based on it */
 880             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 881             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 882             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 883
 884             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 885             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 886             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 887
 888             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 889             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 890             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 891
 892             /* Load parameters for j particles */
 893             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 894                                                                  charge+jnrC+0,charge+jnrD+0,
 895                                                                  charge+jnrE+0,charge+jnrF+0,
 896                                                                  charge+jnrG+0,charge+jnrH+0);
 897
 898             fjx0             = _mm256_setzero_ps();
 899             fjy0             = _mm256_setzero_ps();
 900             fjz0             = _mm256_setzero_ps();
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 907             {
 908
 909             r00              = _mm256_mul_ps(rsq00,rinv00);
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq00             = _mm256_mul_ps(iq0,jq0);
 913
 914             /* EWALD ELECTROSTATICS */
 915
 916             /* Analytical PME correction */
 917             zeta2            = _mm256_mul_ps(beta2,rsq00);
 918             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 919             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 920             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 921             felec            = _mm256_mul_ps(qq00,felec);
 922
 923             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 924
 925             fscal            = felec;
 926
 927             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm256_mul_ps(fscal,dx00);
 931             ty               = _mm256_mul_ps(fscal,dy00);
 932             tz               = _mm256_mul_ps(fscal,dz00);
 933
 934             /* Update vectorial force */
 935             fix0             = _mm256_add_ps(fix0,tx);
 936             fiy0             = _mm256_add_ps(fiy0,ty);
 937             fiz0             = _mm256_add_ps(fiz0,tz);
 938
 939             fjx0             = _mm256_add_ps(fjx0,tx);
 940             fjy0             = _mm256_add_ps(fjy0,ty);
 941             fjz0             = _mm256_add_ps(fjz0,tz);
 942
 943             }
 944
 945             /**************************
 946              * CALCULATE INTERACTIONS *
 947              **************************/
 948
 949             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 950             {
 951
 952             r10              = _mm256_mul_ps(rsq10,rinv10);
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm256_mul_ps(iq1,jq0);
 956
 957             /* EWALD ELECTROSTATICS */
 958
 959             /* Analytical PME correction */
 960             zeta2            = _mm256_mul_ps(beta2,rsq10);
 961             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 962             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 963             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 964             felec            = _mm256_mul_ps(qq10,felec);
 965
 966             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 967
 968             fscal            = felec;
 969
 970             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 971
 972             /* Calculate temporary vectorial force */
 973             tx               = _mm256_mul_ps(fscal,dx10);
 974             ty               = _mm256_mul_ps(fscal,dy10);
 975             tz               = _mm256_mul_ps(fscal,dz10);
 976
 977             /* Update vectorial force */
 978             fix1             = _mm256_add_ps(fix1,tx);
 979             fiy1             = _mm256_add_ps(fiy1,ty);
 980             fiz1             = _mm256_add_ps(fiz1,tz);
 981
 982             fjx0             = _mm256_add_ps(fjx0,tx);
 983             fjy0             = _mm256_add_ps(fjy0,ty);
 984             fjz0             = _mm256_add_ps(fjz0,tz);
 985
 986             }
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 993             {
 994
 995             r20              = _mm256_mul_ps(rsq20,rinv20);
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq20             = _mm256_mul_ps(iq2,jq0);
 999
1000             /* EWALD ELECTROSTATICS */
1001
1002             /* Analytical PME correction */
1003             zeta2            = _mm256_mul_ps(beta2,rsq20);
1004             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1005             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1006             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1007             felec            = _mm256_mul_ps(qq20,felec);
1008
1009             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1010
1011             fscal            = felec;
1012
1013             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1014
1015             /* Calculate temporary vectorial force */
1016             tx               = _mm256_mul_ps(fscal,dx20);
1017             ty               = _mm256_mul_ps(fscal,dy20);
1018             tz               = _mm256_mul_ps(fscal,dz20);
1019
1020             /* Update vectorial force */
1021             fix2             = _mm256_add_ps(fix2,tx);
1022             fiy2             = _mm256_add_ps(fiy2,ty);
1023             fiz2             = _mm256_add_ps(fiz2,tz);
1024
1025             fjx0             = _mm256_add_ps(fjx0,tx);
1026             fjy0             = _mm256_add_ps(fjy0,ty);
1027             fjz0             = _mm256_add_ps(fjz0,tz);
1028
1029             }
1030
1031             fjptrA             = f+j_coord_offsetA;
1032             fjptrB             = f+j_coord_offsetB;
1033             fjptrC             = f+j_coord_offsetC;
1034             fjptrD             = f+j_coord_offsetD;
1035             fjptrE             = f+j_coord_offsetE;
1036             fjptrF             = f+j_coord_offsetF;
1037             fjptrG             = f+j_coord_offsetG;
1038             fjptrH             = f+j_coord_offsetH;
1039
1040             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1041
1042             /* Inner loop uses 180 flops */
1043         }
1044
1045         if(jidx<j_index_end)
1046         {
1047
1048             /* Get j neighbor index, and coordinate index */
1049             jnrlistA         = jjnr[jidx];
1050             jnrlistB         = jjnr[jidx+1];
1051             jnrlistC         = jjnr[jidx+2];
1052             jnrlistD         = jjnr[jidx+3];
1053             jnrlistE         = jjnr[jidx+4];
1054             jnrlistF         = jjnr[jidx+5];
1055             jnrlistG         = jjnr[jidx+6];
1056             jnrlistH         = jjnr[jidx+7];
1057             /* Sign of each element will be negative for non-real atoms.
1058              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1059              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1060              */
1061             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1062                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1063
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1069             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1070             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1071             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1072             j_coord_offsetA  = DIM*jnrA;
1073             j_coord_offsetB  = DIM*jnrB;
1074             j_coord_offsetC  = DIM*jnrC;
1075             j_coord_offsetD  = DIM*jnrD;
1076             j_coord_offsetE  = DIM*jnrE;
1077             j_coord_offsetF  = DIM*jnrF;
1078             j_coord_offsetG  = DIM*jnrG;
1079             j_coord_offsetH  = DIM*jnrH;
1080
1081             /* load j atom coordinates */
1082             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1083                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1084                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1085                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1086                                                  &jx0,&jy0,&jz0);
1087
1088             /* Calculate displacement vector */
1089             dx00             = _mm256_sub_ps(ix0,jx0);
1090             dy00             = _mm256_sub_ps(iy0,jy0);
1091             dz00             = _mm256_sub_ps(iz0,jz0);
1092             dx10             = _mm256_sub_ps(ix1,jx0);
1093             dy10             = _mm256_sub_ps(iy1,jy0);
1094             dz10             = _mm256_sub_ps(iz1,jz0);
1095             dx20             = _mm256_sub_ps(ix2,jx0);
1096             dy20             = _mm256_sub_ps(iy2,jy0);
1097             dz20             = _mm256_sub_ps(iz2,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1101             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1102             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1103
1104             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1105             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1106             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1107
1108             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1109             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1110             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1111
1112             /* Load parameters for j particles */
1113             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1114                                                                  charge+jnrC+0,charge+jnrD+0,
1115                                                                  charge+jnrE+0,charge+jnrF+0,
1116                                                                  charge+jnrG+0,charge+jnrH+0);
1117
1118             fjx0             = _mm256_setzero_ps();
1119             fjy0             = _mm256_setzero_ps();
1120             fjz0             = _mm256_setzero_ps();
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1127             {
1128
1129             r00              = _mm256_mul_ps(rsq00,rinv00);
1130             r00              = _mm256_andnot_ps(dummy_mask,r00);
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             qq00             = _mm256_mul_ps(iq0,jq0);
1134
1135             /* EWALD ELECTROSTATICS */
1136
1137             /* Analytical PME correction */
1138             zeta2            = _mm256_mul_ps(beta2,rsq00);
1139             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1140             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1141             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1142             felec            = _mm256_mul_ps(qq00,felec);
1143
1144             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1149
1150             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm256_mul_ps(fscal,dx00);
1154             ty               = _mm256_mul_ps(fscal,dy00);
1155             tz               = _mm256_mul_ps(fscal,dz00);
1156
1157             /* Update vectorial force */
1158             fix0             = _mm256_add_ps(fix0,tx);
1159             fiy0             = _mm256_add_ps(fiy0,ty);
1160             fiz0             = _mm256_add_ps(fiz0,tz);
1161
1162             fjx0             = _mm256_add_ps(fjx0,tx);
1163             fjy0             = _mm256_add_ps(fjy0,ty);
1164             fjz0             = _mm256_add_ps(fjz0,tz);
1165
1166             }
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1173             {
1174
1175             r10              = _mm256_mul_ps(rsq10,rinv10);
1176             r10              = _mm256_andnot_ps(dummy_mask,r10);
1177
1178             /* Compute parameters for interactions between i and j atoms */
1179             qq10             = _mm256_mul_ps(iq1,jq0);
1180
1181             /* EWALD ELECTROSTATICS */
1182
1183             /* Analytical PME correction */
1184             zeta2            = _mm256_mul_ps(beta2,rsq10);
1185             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1186             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1187             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1188             felec            = _mm256_mul_ps(qq10,felec);
1189
1190             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1195
1196             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_ps(fscal,dx10);
1200             ty               = _mm256_mul_ps(fscal,dy10);
1201             tz               = _mm256_mul_ps(fscal,dz10);
1202
1203             /* Update vectorial force */
1204             fix1             = _mm256_add_ps(fix1,tx);
1205             fiy1             = _mm256_add_ps(fiy1,ty);
1206             fiz1             = _mm256_add_ps(fiz1,tz);
1207
1208             fjx0             = _mm256_add_ps(fjx0,tx);
1209             fjy0             = _mm256_add_ps(fjy0,ty);
1210             fjz0             = _mm256_add_ps(fjz0,tz);
1211
1212             }
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1219             {
1220
1221             r20              = _mm256_mul_ps(rsq20,rinv20);
1222             r20              = _mm256_andnot_ps(dummy_mask,r20);
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             qq20             = _mm256_mul_ps(iq2,jq0);
1226
1227             /* EWALD ELECTROSTATICS */
1228
1229             /* Analytical PME correction */
1230             zeta2            = _mm256_mul_ps(beta2,rsq20);
1231             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1232             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1233             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1234             felec            = _mm256_mul_ps(qq20,felec);
1235
1236             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1241
1242             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm256_mul_ps(fscal,dx20);
1246             ty               = _mm256_mul_ps(fscal,dy20);
1247             tz               = _mm256_mul_ps(fscal,dz20);
1248
1249             /* Update vectorial force */
1250             fix2             = _mm256_add_ps(fix2,tx);
1251             fiy2             = _mm256_add_ps(fiy2,ty);
1252             fiz2             = _mm256_add_ps(fiz2,tz);
1253
1254             fjx0             = _mm256_add_ps(fjx0,tx);
1255             fjy0             = _mm256_add_ps(fjy0,ty);
1256             fjz0             = _mm256_add_ps(fjz0,tz);
1257
1258             }
1259
1260             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1261             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1262             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1263             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1264             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1265             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1266             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1267             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1268
1269             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1270
1271             /* Inner loop uses 183 flops */
1272         }
1273
1274         /* End of innermost loop */
1275
1276         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1277                                                  f+i_coord_offset,fshift+i_shift_offset);
1278
1279         /* Increment number of inner iterations */
1280         inneriter                  += j_index_end - j_index_start;
1281
1282         /* Outer loop uses 18 flops */
1283     }
1284
1285     /* Increment number of outer iterations */
1286     outeriter        += nri;
1287
1288     /* Update outer/inner flops */
1289
1290     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*183);
1291 }