src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwNone_GeomW3P1_avx_256_double.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_double 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_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m128i          ewitab;
 100     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 102     real             *ewtab;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(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_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 123     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 124     beta2            = _mm256_mul_pd(beta,beta);
 125     beta3            = _mm256_mul_pd(beta,beta2);
 126
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 134     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 135     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm256_setzero_pd();
 171         fiy0             = _mm256_setzero_pd();
 172         fiz0             = _mm256_setzero_pd();
 173         fix1             = _mm256_setzero_pd();
 174         fiy1             = _mm256_setzero_pd();
 175         fiz1             = _mm256_setzero_pd();
 176         fix2             = _mm256_setzero_pd();
 177         fiy2             = _mm256_setzero_pd();
 178         fiz2             = _mm256_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm256_sub_pd(ix0,jx0);
 204             dy00             = _mm256_sub_pd(iy0,jy0);
 205             dz00             = _mm256_sub_pd(iz0,jz0);
 206             dx10             = _mm256_sub_pd(ix1,jx0);
 207             dy10             = _mm256_sub_pd(iy1,jy0);
 208             dz10             = _mm256_sub_pd(iz1,jz0);
 209             dx20             = _mm256_sub_pd(ix2,jx0);
 210             dy20             = _mm256_sub_pd(iy2,jy0);
 211             dz20             = _mm256_sub_pd(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 219             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 220             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 221
 222             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0);
 229
 230             fjx0             = _mm256_setzero_pd();
 231             fjy0             = _mm256_setzero_pd();
 232             fjz0             = _mm256_setzero_pd();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm256_mul_pd(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm256_mul_pd(iq0,jq0);
 242
 243             /* EWALD ELECTROSTATICS */
 244
 245             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 246             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 247             ewitab           = _mm256_cvttpd_epi32(ewrt);
 248             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 249             ewitab           = _mm_slli_epi32(ewitab,2);
 250             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 251             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 252             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 253             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 254             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 255             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 256             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 257             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 258             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velecsum         = _mm256_add_pd(velecsum,velec);
 262
 263             fscal            = felec;
 264
 265             /* Calculate temporary vectorial force */
 266             tx               = _mm256_mul_pd(fscal,dx00);
 267             ty               = _mm256_mul_pd(fscal,dy00);
 268             tz               = _mm256_mul_pd(fscal,dz00);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm256_add_pd(fix0,tx);
 272             fiy0             = _mm256_add_pd(fiy0,ty);
 273             fiz0             = _mm256_add_pd(fiz0,tz);
 274
 275             fjx0             = _mm256_add_pd(fjx0,tx);
 276             fjy0             = _mm256_add_pd(fjy0,ty);
 277             fjz0             = _mm256_add_pd(fjz0,tz);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r10              = _mm256_mul_pd(rsq10,rinv10);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm256_mul_pd(iq1,jq0);
 287
 288             /* EWALD ELECTROSTATICS */
 289
 290             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 291             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 292             ewitab           = _mm256_cvttpd_epi32(ewrt);
 293             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 294             ewitab           = _mm_slli_epi32(ewitab,2);
 295             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 296             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 297             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 298             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 299             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 300             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 301             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 302             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 303             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velecsum         = _mm256_add_pd(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_pd(fscal,dx10);
 312             ty               = _mm256_mul_pd(fscal,dy10);
 313             tz               = _mm256_mul_pd(fscal,dz10);
 314
 315             /* Update vectorial force */
 316             fix1             = _mm256_add_pd(fix1,tx);
 317             fiy1             = _mm256_add_pd(fiy1,ty);
 318             fiz1             = _mm256_add_pd(fiz1,tz);
 319
 320             fjx0             = _mm256_add_pd(fjx0,tx);
 321             fjy0             = _mm256_add_pd(fjy0,ty);
 322             fjz0             = _mm256_add_pd(fjz0,tz);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             r20              = _mm256_mul_pd(rsq20,rinv20);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq20             = _mm256_mul_pd(iq2,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 336             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 337             ewitab           = _mm256_cvttpd_epi32(ewrt);
 338             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 339             ewitab           = _mm_slli_epi32(ewitab,2);
 340             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 341             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 342             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 343             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 344             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 345             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 346             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 347             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 348             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm256_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx20);
 357             ty               = _mm256_mul_pd(fscal,dy20);
 358             tz               = _mm256_mul_pd(fscal,dz20);
 359
 360             /* Update vectorial force */
 361             fix2             = _mm256_add_pd(fix2,tx);
 362             fiy2             = _mm256_add_pd(fiy2,ty);
 363             fiz2             = _mm256_add_pd(fiz2,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             fjptrA             = f+j_coord_offsetA;
 370             fjptrB             = f+j_coord_offsetB;
 371             fjptrC             = f+j_coord_offsetC;
 372             fjptrD             = f+j_coord_offsetD;
 373
 374             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 126 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             /* Get j neighbor index, and coordinate index */
 383             jnrlistA         = jjnr[jidx];
 384             jnrlistB         = jjnr[jidx+1];
 385             jnrlistC         = jjnr[jidx+2];
 386             jnrlistD         = jjnr[jidx+3];
 387             /* Sign of each element will be negative for non-real atoms.
 388              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 389              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 390              */
 391             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 392
 393             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 394             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 395             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 396
 397             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 398             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 399             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 400             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 401             j_coord_offsetA  = DIM*jnrA;
 402             j_coord_offsetB  = DIM*jnrB;
 403             j_coord_offsetC  = DIM*jnrC;
 404             j_coord_offsetD  = DIM*jnrD;
 405
 406             /* load j atom coordinates */
 407             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 408                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 409                                                  &jx0,&jy0,&jz0);
 410
 411             /* Calculate displacement vector */
 412             dx00             = _mm256_sub_pd(ix0,jx0);
 413             dy00             = _mm256_sub_pd(iy0,jy0);
 414             dz00             = _mm256_sub_pd(iz0,jz0);
 415             dx10             = _mm256_sub_pd(ix1,jx0);
 416             dy10             = _mm256_sub_pd(iy1,jy0);
 417             dz10             = _mm256_sub_pd(iz1,jz0);
 418             dx20             = _mm256_sub_pd(ix2,jx0);
 419             dy20             = _mm256_sub_pd(iy2,jy0);
 420             dz20             = _mm256_sub_pd(iz2,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 424             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 425             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 426
 427             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 428             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 429             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 430
 431             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 432             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 433             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 434
 435             /* Load parameters for j particles */
 436             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 437                                                                  charge+jnrC+0,charge+jnrD+0);
 438
 439             fjx0             = _mm256_setzero_pd();
 440             fjy0             = _mm256_setzero_pd();
 441             fjz0             = _mm256_setzero_pd();
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             r00              = _mm256_mul_pd(rsq00,rinv00);
 448             r00              = _mm256_andnot_pd(dummy_mask,r00);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq00             = _mm256_mul_pd(iq0,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 457             ewitab           = _mm256_cvttpd_epi32(ewrt);
 458             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 462             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 463             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 464             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 465             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 466             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 467             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 468             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm256_andnot_pd(dummy_mask,velec);
 472             velecsum         = _mm256_add_pd(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm256_mul_pd(fscal,dx00);
 480             ty               = _mm256_mul_pd(fscal,dy00);
 481             tz               = _mm256_mul_pd(fscal,dz00);
 482
 483             /* Update vectorial force */
 484             fix0             = _mm256_add_pd(fix0,tx);
 485             fiy0             = _mm256_add_pd(fiy0,ty);
 486             fiz0             = _mm256_add_pd(fiz0,tz);
 487
 488             fjx0             = _mm256_add_pd(fjx0,tx);
 489             fjy0             = _mm256_add_pd(fjy0,ty);
 490             fjz0             = _mm256_add_pd(fjz0,tz);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r10              = _mm256_mul_pd(rsq10,rinv10);
 497             r10              = _mm256_andnot_pd(dummy_mask,r10);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm256_mul_pd(iq1,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 506             ewitab           = _mm256_cvttpd_epi32(ewrt);
 507             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 508             ewitab           = _mm_slli_epi32(ewitab,2);
 509             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 510             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 511             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 512             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 513             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 514             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 515             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 516             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 517             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm256_andnot_pd(dummy_mask,velec);
 521             velecsum         = _mm256_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm256_mul_pd(fscal,dx10);
 529             ty               = _mm256_mul_pd(fscal,dy10);
 530             tz               = _mm256_mul_pd(fscal,dz10);
 531
 532             /* Update vectorial force */
 533             fix1             = _mm256_add_pd(fix1,tx);
 534             fiy1             = _mm256_add_pd(fiy1,ty);
 535             fiz1             = _mm256_add_pd(fiz1,tz);
 536
 537             fjx0             = _mm256_add_pd(fjx0,tx);
 538             fjy0             = _mm256_add_pd(fjy0,ty);
 539             fjz0             = _mm256_add_pd(fjz0,tz);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             r20              = _mm256_mul_pd(rsq20,rinv20);
 546             r20              = _mm256_andnot_pd(dummy_mask,r20);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq20             = _mm256_mul_pd(iq2,jq0);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 555             ewitab           = _mm256_cvttpd_epi32(ewrt);
 556             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 560             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 561             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 562             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 563             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 564             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 565             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 566             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm256_andnot_pd(dummy_mask,velec);
 570             velecsum         = _mm256_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm256_mul_pd(fscal,dx20);
 578             ty               = _mm256_mul_pd(fscal,dy20);
 579             tz               = _mm256_mul_pd(fscal,dz20);
 580
 581             /* Update vectorial force */
 582             fix2             = _mm256_add_pd(fix2,tx);
 583             fiy2             = _mm256_add_pd(fiy2,ty);
 584             fiz2             = _mm256_add_pd(fiz2,tz);
 585
 586             fjx0             = _mm256_add_pd(fjx0,tx);
 587             fjy0             = _mm256_add_pd(fjy0,ty);
 588             fjz0             = _mm256_add_pd(fjz0,tz);
 589
 590             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 591             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 592             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 593             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 594
 595             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 596
 597             /* Inner loop uses 129 flops */
 598         }
 599
 600         /* End of innermost loop */
 601
 602         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 603                                                  f+i_coord_offset,fshift+i_shift_offset);
 604
 605         ggid                        = gid[iidx];
 606         /* Update potential energies */
 607         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 19 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*129);
 621 }
 622 /*
 623  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_double
 624  * Electrostatics interaction: Ewald
 625  * VdW interaction:            None
 626  * Geometry:                   Water3-Particle
 627  * Calculate force/pot:        Force
 628  */
 629 void
 630 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_double
 631                     (t_nblist                    * gmx_restrict       nlist,
 632                      rvec                        * gmx_restrict          xx,
 633                      rvec                        * gmx_restrict          ff,
 634                      t_forcerec                  * gmx_restrict          fr,
 635                      t_mdatoms                   * gmx_restrict     mdatoms,
 636                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 637                      t_nrnb                      * gmx_restrict        nrnb)
 638 {
 639     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 640      * just 0 for non-waters.
 641      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 642      * jnr indices corresponding to data put in the four positions in the SIMD register.
 643      */
 644     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 645     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 646     int              jnrA,jnrB,jnrC,jnrD;
 647     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 648     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 649     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 650     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 651     real             rcutoff_scalar;
 652     real             *shiftvec,*fshift,*x,*f;
 653     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 654     real             scratch[4*DIM];
 655     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 656     real *           vdwioffsetptr0;
 657     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 658     real *           vdwioffsetptr1;
 659     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     real *           vdwioffsetptr2;
 661     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 663     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 668     real             *charge;
 669     __m128i          ewitab;
 670     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 671     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 672     real             *ewtab;
 673     __m256d          dummy_mask,cutoff_mask;
 674     __m128           tmpmask0,tmpmask1;
 675     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 676     __m256d          one     = _mm256_set1_pd(1.0);
 677     __m256d          two     = _mm256_set1_pd(2.0);
 678     x                = xx[0];
 679     f                = ff[0];
 680
 681     nri              = nlist->nri;
 682     iinr             = nlist->iinr;
 683     jindex           = nlist->jindex;
 684     jjnr             = nlist->jjnr;
 685     shiftidx         = nlist->shift;
 686     gid              = nlist->gid;
 687     shiftvec         = fr->shift_vec[0];
 688     fshift           = fr->fshift[0];
 689     facel            = _mm256_set1_pd(fr->epsfac);
 690     charge           = mdatoms->chargeA;
 691
 692     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 693     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 694     beta2            = _mm256_mul_pd(beta,beta);
 695     beta3            = _mm256_mul_pd(beta,beta2);
 696
 697     ewtab            = fr->ic->tabq_coul_F;
 698     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 699     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 700
 701     /* Setup water-specific parameters */
 702     inr              = nlist->iinr[0];
 703     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 704     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 705     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 706
 707     /* Avoid stupid compiler warnings */
 708     jnrA = jnrB = jnrC = jnrD = 0;
 709     j_coord_offsetA = 0;
 710     j_coord_offsetB = 0;
 711     j_coord_offsetC = 0;
 712     j_coord_offsetD = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     for(iidx=0;iidx<4*DIM;iidx++)
 718     {
 719         scratch[iidx] = 0.0;
 720     }
 721
 722     /* Start outer loop over neighborlists */
 723     for(iidx=0; iidx<nri; iidx++)
 724     {
 725         /* Load shift vector for this list */
 726         i_shift_offset   = DIM*shiftidx[iidx];
 727
 728         /* Load limits for loop over neighbors */
 729         j_index_start    = jindex[iidx];
 730         j_index_end      = jindex[iidx+1];
 731
 732         /* Get outer coordinate index */
 733         inr              = iinr[iidx];
 734         i_coord_offset   = DIM*inr;
 735
 736         /* Load i particle coords and add shift vector */
 737         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 738                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 739
 740         fix0             = _mm256_setzero_pd();
 741         fiy0             = _mm256_setzero_pd();
 742         fiz0             = _mm256_setzero_pd();
 743         fix1             = _mm256_setzero_pd();
 744         fiy1             = _mm256_setzero_pd();
 745         fiz1             = _mm256_setzero_pd();
 746         fix2             = _mm256_setzero_pd();
 747         fiy2             = _mm256_setzero_pd();
 748         fiz2             = _mm256_setzero_pd();
 749
 750         /* Start inner kernel loop */
 751         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 752         {
 753
 754             /* Get j neighbor index, and coordinate index */
 755             jnrA             = jjnr[jidx];
 756             jnrB             = jjnr[jidx+1];
 757             jnrC             = jjnr[jidx+2];
 758             jnrD             = jjnr[jidx+3];
 759             j_coord_offsetA  = DIM*jnrA;
 760             j_coord_offsetB  = DIM*jnrB;
 761             j_coord_offsetC  = DIM*jnrC;
 762             j_coord_offsetD  = DIM*jnrD;
 763
 764             /* load j atom coordinates */
 765             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 766                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 767                                                  &jx0,&jy0,&jz0);
 768
 769             /* Calculate displacement vector */
 770             dx00             = _mm256_sub_pd(ix0,jx0);
 771             dy00             = _mm256_sub_pd(iy0,jy0);
 772             dz00             = _mm256_sub_pd(iz0,jz0);
 773             dx10             = _mm256_sub_pd(ix1,jx0);
 774             dy10             = _mm256_sub_pd(iy1,jy0);
 775             dz10             = _mm256_sub_pd(iz1,jz0);
 776             dx20             = _mm256_sub_pd(ix2,jx0);
 777             dy20             = _mm256_sub_pd(iy2,jy0);
 778             dz20             = _mm256_sub_pd(iz2,jz0);
 779
 780             /* Calculate squared distance and things based on it */
 781             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 782             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 783             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 784
 785             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 786             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 787             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 788
 789             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 790             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 791             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 792
 793             /* Load parameters for j particles */
 794             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 795                                                                  charge+jnrC+0,charge+jnrD+0);
 796
 797             fjx0             = _mm256_setzero_pd();
 798             fjy0             = _mm256_setzero_pd();
 799             fjz0             = _mm256_setzero_pd();
 800
 801             /**************************
 802              * CALCULATE INTERACTIONS *
 803              **************************/
 804
 805             r00              = _mm256_mul_pd(rsq00,rinv00);
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq00             = _mm256_mul_pd(iq0,jq0);
 809
 810             /* EWALD ELECTROSTATICS */
 811
 812             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 813             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 814             ewitab           = _mm256_cvttpd_epi32(ewrt);
 815             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 816             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 817                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 818                                             &ewtabF,&ewtabFn);
 819             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 820             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 821
 822             fscal            = felec;
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm256_mul_pd(fscal,dx00);
 826             ty               = _mm256_mul_pd(fscal,dy00);
 827             tz               = _mm256_mul_pd(fscal,dz00);
 828
 829             /* Update vectorial force */
 830             fix0             = _mm256_add_pd(fix0,tx);
 831             fiy0             = _mm256_add_pd(fiy0,ty);
 832             fiz0             = _mm256_add_pd(fiz0,tz);
 833
 834             fjx0             = _mm256_add_pd(fjx0,tx);
 835             fjy0             = _mm256_add_pd(fjy0,ty);
 836             fjz0             = _mm256_add_pd(fjz0,tz);
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             r10              = _mm256_mul_pd(rsq10,rinv10);
 843
 844             /* Compute parameters for interactions between i and j atoms */
 845             qq10             = _mm256_mul_pd(iq1,jq0);
 846
 847             /* EWALD ELECTROSTATICS */
 848
 849             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 850             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 851             ewitab           = _mm256_cvttpd_epi32(ewrt);
 852             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 853             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 854                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 855                                             &ewtabF,&ewtabFn);
 856             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 857             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 858
 859             fscal            = felec;
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm256_mul_pd(fscal,dx10);
 863             ty               = _mm256_mul_pd(fscal,dy10);
 864             tz               = _mm256_mul_pd(fscal,dz10);
 865
 866             /* Update vectorial force */
 867             fix1             = _mm256_add_pd(fix1,tx);
 868             fiy1             = _mm256_add_pd(fiy1,ty);
 869             fiz1             = _mm256_add_pd(fiz1,tz);
 870
 871             fjx0             = _mm256_add_pd(fjx0,tx);
 872             fjy0             = _mm256_add_pd(fjy0,ty);
 873             fjz0             = _mm256_add_pd(fjz0,tz);
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             r20              = _mm256_mul_pd(rsq20,rinv20);
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq20             = _mm256_mul_pd(iq2,jq0);
 883
 884             /* EWALD ELECTROSTATICS */
 885
 886             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 887             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 888             ewitab           = _mm256_cvttpd_epi32(ewrt);
 889             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 890             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 891                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 892                                             &ewtabF,&ewtabFn);
 893             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 894             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 895
 896             fscal            = felec;
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm256_mul_pd(fscal,dx20);
 900             ty               = _mm256_mul_pd(fscal,dy20);
 901             tz               = _mm256_mul_pd(fscal,dz20);
 902
 903             /* Update vectorial force */
 904             fix2             = _mm256_add_pd(fix2,tx);
 905             fiy2             = _mm256_add_pd(fiy2,ty);
 906             fiz2             = _mm256_add_pd(fiz2,tz);
 907
 908             fjx0             = _mm256_add_pd(fjx0,tx);
 909             fjy0             = _mm256_add_pd(fjy0,ty);
 910             fjz0             = _mm256_add_pd(fjz0,tz);
 911
 912             fjptrA             = f+j_coord_offsetA;
 913             fjptrB             = f+j_coord_offsetB;
 914             fjptrC             = f+j_coord_offsetC;
 915             fjptrD             = f+j_coord_offsetD;
 916
 917             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 918
 919             /* Inner loop uses 111 flops */
 920         }
 921
 922         if(jidx<j_index_end)
 923         {
 924
 925             /* Get j neighbor index, and coordinate index */
 926             jnrlistA         = jjnr[jidx];
 927             jnrlistB         = jjnr[jidx+1];
 928             jnrlistC         = jjnr[jidx+2];
 929             jnrlistD         = jjnr[jidx+3];
 930             /* Sign of each element will be negative for non-real atoms.
 931              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 932              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 933              */
 934             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 935
 936             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 937             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 938             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 939
 940             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 941             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 942             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 943             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 944             j_coord_offsetA  = DIM*jnrA;
 945             j_coord_offsetB  = DIM*jnrB;
 946             j_coord_offsetC  = DIM*jnrC;
 947             j_coord_offsetD  = DIM*jnrD;
 948
 949             /* load j atom coordinates */
 950             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 951                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 952                                                  &jx0,&jy0,&jz0);
 953
 954             /* Calculate displacement vector */
 955             dx00             = _mm256_sub_pd(ix0,jx0);
 956             dy00             = _mm256_sub_pd(iy0,jy0);
 957             dz00             = _mm256_sub_pd(iz0,jz0);
 958             dx10             = _mm256_sub_pd(ix1,jx0);
 959             dy10             = _mm256_sub_pd(iy1,jy0);
 960             dz10             = _mm256_sub_pd(iz1,jz0);
 961             dx20             = _mm256_sub_pd(ix2,jx0);
 962             dy20             = _mm256_sub_pd(iy2,jy0);
 963             dz20             = _mm256_sub_pd(iz2,jz0);
 964
 965             /* Calculate squared distance and things based on it */
 966             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 967             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 968             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 969
 970             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 971             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 972             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 973
 974             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 975             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 976             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 977
 978             /* Load parameters for j particles */
 979             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 980                                                                  charge+jnrC+0,charge+jnrD+0);
 981
 982             fjx0             = _mm256_setzero_pd();
 983             fjy0             = _mm256_setzero_pd();
 984             fjz0             = _mm256_setzero_pd();
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             r00              = _mm256_mul_pd(rsq00,rinv00);
 991             r00              = _mm256_andnot_pd(dummy_mask,r00);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq00             = _mm256_mul_pd(iq0,jq0);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1000             ewitab           = _mm256_cvttpd_epi32(ewrt);
1001             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1002             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1003                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1004                                             &ewtabF,&ewtabFn);
1005             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1006             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm256_mul_pd(fscal,dx00);
1014             ty               = _mm256_mul_pd(fscal,dy00);
1015             tz               = _mm256_mul_pd(fscal,dz00);
1016
1017             /* Update vectorial force */
1018             fix0             = _mm256_add_pd(fix0,tx);
1019             fiy0             = _mm256_add_pd(fiy0,ty);
1020             fiz0             = _mm256_add_pd(fiz0,tz);
1021
1022             fjx0             = _mm256_add_pd(fjx0,tx);
1023             fjy0             = _mm256_add_pd(fjy0,ty);
1024             fjz0             = _mm256_add_pd(fjz0,tz);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             r10              = _mm256_mul_pd(rsq10,rinv10);
1031             r10              = _mm256_andnot_pd(dummy_mask,r10);
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq10             = _mm256_mul_pd(iq1,jq0);
1035
1036             /* EWALD ELECTROSTATICS */
1037
1038             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1039             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1040             ewitab           = _mm256_cvttpd_epi32(ewrt);
1041             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1042             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1043                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1044                                             &ewtabF,&ewtabFn);
1045             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1046             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm256_mul_pd(fscal,dx10);
1054             ty               = _mm256_mul_pd(fscal,dy10);
1055             tz               = _mm256_mul_pd(fscal,dz10);
1056
1057             /* Update vectorial force */
1058             fix1             = _mm256_add_pd(fix1,tx);
1059             fiy1             = _mm256_add_pd(fiy1,ty);
1060             fiz1             = _mm256_add_pd(fiz1,tz);
1061
1062             fjx0             = _mm256_add_pd(fjx0,tx);
1063             fjy0             = _mm256_add_pd(fjy0,ty);
1064             fjz0             = _mm256_add_pd(fjz0,tz);
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             r20              = _mm256_mul_pd(rsq20,rinv20);
1071             r20              = _mm256_andnot_pd(dummy_mask,r20);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq20             = _mm256_mul_pd(iq2,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1080             ewitab           = _mm256_cvttpd_epi32(ewrt);
1081             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1082             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1083                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1084                                             &ewtabF,&ewtabFn);
1085             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1086             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_pd(fscal,dx20);
1094             ty               = _mm256_mul_pd(fscal,dy20);
1095             tz               = _mm256_mul_pd(fscal,dz20);
1096
1097             /* Update vectorial force */
1098             fix2             = _mm256_add_pd(fix2,tx);
1099             fiy2             = _mm256_add_pd(fiy2,ty);
1100             fiz2             = _mm256_add_pd(fiz2,tz);
1101
1102             fjx0             = _mm256_add_pd(fjx0,tx);
1103             fjy0             = _mm256_add_pd(fjy0,ty);
1104             fjz0             = _mm256_add_pd(fjz0,tz);
1105
1106             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1107             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1108             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1109             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1110
1111             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 114 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1119                                                  f+i_coord_offset,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*114);
1133 }