src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwNone_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr1;
  84     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     real *           vdwioffsetptr2;
  86     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     real *           vdwioffsetptr3;
  88     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          ewitab;
  97     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256d          dummy_mask,cutoff_mask;
 101     __m128           tmpmask0,tmpmask1;
 102     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 103     __m256d          one     = _mm256_set1_pd(1.0);
 104     __m256d          two     = _mm256_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_pd(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118
 119     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 120     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 121     beta2            = _mm256_mul_pd(beta,beta);
 122     beta3            = _mm256_mul_pd(beta,beta2);
 123
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 131     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 132     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 165                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix1             = _mm256_setzero_pd();
 168         fiy1             = _mm256_setzero_pd();
 169         fiz1             = _mm256_setzero_pd();
 170         fix2             = _mm256_setzero_pd();
 171         fiy2             = _mm256_setzero_pd();
 172         fiz2             = _mm256_setzero_pd();
 173         fix3             = _mm256_setzero_pd();
 174         fiy3             = _mm256_setzero_pd();
 175         fiz3             = _mm256_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm256_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 197                                                  &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx10             = _mm256_sub_pd(ix1,jx0);
 201             dy10             = _mm256_sub_pd(iy1,jy0);
 202             dz10             = _mm256_sub_pd(iz1,jz0);
 203             dx20             = _mm256_sub_pd(ix2,jx0);
 204             dy20             = _mm256_sub_pd(iy2,jy0);
 205             dz20             = _mm256_sub_pd(iz2,jz0);
 206             dx30             = _mm256_sub_pd(ix3,jx0);
 207             dy30             = _mm256_sub_pd(iy3,jy0);
 208             dz30             = _mm256_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 213             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 214
 215             rinv10           = avx256_invsqrt_d(rsq10);
 216             rinv20           = avx256_invsqrt_d(rsq20);
 217             rinv30           = avx256_invsqrt_d(rsq30);
 218
 219             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 221             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 225                                                                  charge+jnrC+0,charge+jnrD+0);
 226
 227             fjx0             = _mm256_setzero_pd();
 228             fjy0             = _mm256_setzero_pd();
 229             fjz0             = _mm256_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             r10              = _mm256_mul_pd(rsq10,rinv10);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq10             = _mm256_mul_pd(iq1,jq0);
 239
 240             /* EWALD ELECTROSTATICS */
 241
 242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 243             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 244             ewitab           = _mm256_cvttpd_epi32(ewrt);
 245             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 246             ewitab           = _mm_slli_epi32(ewitab,2);
 247             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 248             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 249             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 250             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 251             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 252             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 253             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 254             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 255             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velecsum         = _mm256_add_pd(velecsum,velec);
 259
 260             fscal            = felec;
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm256_mul_pd(fscal,dx10);
 264             ty               = _mm256_mul_pd(fscal,dy10);
 265             tz               = _mm256_mul_pd(fscal,dz10);
 266
 267             /* Update vectorial force */
 268             fix1             = _mm256_add_pd(fix1,tx);
 269             fiy1             = _mm256_add_pd(fiy1,ty);
 270             fiz1             = _mm256_add_pd(fiz1,tz);
 271
 272             fjx0             = _mm256_add_pd(fjx0,tx);
 273             fjy0             = _mm256_add_pd(fjy0,ty);
 274             fjz0             = _mm256_add_pd(fjz0,tz);
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r20              = _mm256_mul_pd(rsq20,rinv20);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq20             = _mm256_mul_pd(iq2,jq0);
 284
 285             /* EWALD ELECTROSTATICS */
 286
 287             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 288             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 289             ewitab           = _mm256_cvttpd_epi32(ewrt);
 290             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 291             ewitab           = _mm_slli_epi32(ewitab,2);
 292             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 293             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 294             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 295             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 297             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 298             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 299             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 300             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm256_add_pd(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm256_mul_pd(fscal,dx20);
 309             ty               = _mm256_mul_pd(fscal,dy20);
 310             tz               = _mm256_mul_pd(fscal,dz20);
 311
 312             /* Update vectorial force */
 313             fix2             = _mm256_add_pd(fix2,tx);
 314             fiy2             = _mm256_add_pd(fiy2,ty);
 315             fiz2             = _mm256_add_pd(fiz2,tz);
 316
 317             fjx0             = _mm256_add_pd(fjx0,tx);
 318             fjy0             = _mm256_add_pd(fjy0,ty);
 319             fjz0             = _mm256_add_pd(fjz0,tz);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             r30              = _mm256_mul_pd(rsq30,rinv30);
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq30             = _mm256_mul_pd(iq3,jq0);
 329
 330             /* EWALD ELECTROSTATICS */
 331
 332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 333             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 334             ewitab           = _mm256_cvttpd_epi32(ewrt);
 335             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 336             ewitab           = _mm_slli_epi32(ewitab,2);
 337             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 338             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 339             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 340             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 341             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 342             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 343             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 344             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 345             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm256_add_pd(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_pd(fscal,dx30);
 354             ty               = _mm256_mul_pd(fscal,dy30);
 355             tz               = _mm256_mul_pd(fscal,dz30);
 356
 357             /* Update vectorial force */
 358             fix3             = _mm256_add_pd(fix3,tx);
 359             fiy3             = _mm256_add_pd(fiy3,ty);
 360             fiz3             = _mm256_add_pd(fiz3,tz);
 361
 362             fjx0             = _mm256_add_pd(fjx0,tx);
 363             fjy0             = _mm256_add_pd(fjy0,ty);
 364             fjz0             = _mm256_add_pd(fjz0,tz);
 365
 366             fjptrA             = f+j_coord_offsetA;
 367             fjptrB             = f+j_coord_offsetB;
 368             fjptrC             = f+j_coord_offsetC;
 369             fjptrD             = f+j_coord_offsetD;
 370
 371             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 372
 373             /* Inner loop uses 126 flops */
 374         }
 375
 376         if(jidx<j_index_end)
 377         {
 378
 379             /* Get j neighbor index, and coordinate index */
 380             jnrlistA         = jjnr[jidx];
 381             jnrlistB         = jjnr[jidx+1];
 382             jnrlistC         = jjnr[jidx+2];
 383             jnrlistD         = jjnr[jidx+3];
 384             /* Sign of each element will be negative for non-real atoms.
 385              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 386              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 387              */
 388             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 389
 390             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 391             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 392             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 393
 394             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 395             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 396             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 397             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 398             j_coord_offsetA  = DIM*jnrA;
 399             j_coord_offsetB  = DIM*jnrB;
 400             j_coord_offsetC  = DIM*jnrC;
 401             j_coord_offsetD  = DIM*jnrD;
 402
 403             /* load j atom coordinates */
 404             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 405                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 406                                                  &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx10             = _mm256_sub_pd(ix1,jx0);
 410             dy10             = _mm256_sub_pd(iy1,jy0);
 411             dz10             = _mm256_sub_pd(iz1,jz0);
 412             dx20             = _mm256_sub_pd(ix2,jx0);
 413             dy20             = _mm256_sub_pd(iy2,jy0);
 414             dz20             = _mm256_sub_pd(iz2,jz0);
 415             dx30             = _mm256_sub_pd(ix3,jx0);
 416             dy30             = _mm256_sub_pd(iy3,jy0);
 417             dz30             = _mm256_sub_pd(iz3,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 422             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 423
 424             rinv10           = avx256_invsqrt_d(rsq10);
 425             rinv20           = avx256_invsqrt_d(rsq20);
 426             rinv30           = avx256_invsqrt_d(rsq30);
 427
 428             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 429             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 430             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 431
 432             /* Load parameters for j particles */
 433             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 434                                                                  charge+jnrC+0,charge+jnrD+0);
 435
 436             fjx0             = _mm256_setzero_pd();
 437             fjy0             = _mm256_setzero_pd();
 438             fjz0             = _mm256_setzero_pd();
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r10              = _mm256_mul_pd(rsq10,rinv10);
 445             r10              = _mm256_andnot_pd(dummy_mask,r10);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq10             = _mm256_mul_pd(iq1,jq0);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 454             ewitab           = _mm256_cvttpd_epi32(ewrt);
 455             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 456             ewitab           = _mm_slli_epi32(ewitab,2);
 457             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 458             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 459             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 460             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 461             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 462             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 463             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 464             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 465             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm256_andnot_pd(dummy_mask,velec);
 469             velecsum         = _mm256_add_pd(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm256_mul_pd(fscal,dx10);
 477             ty               = _mm256_mul_pd(fscal,dy10);
 478             tz               = _mm256_mul_pd(fscal,dz10);
 479
 480             /* Update vectorial force */
 481             fix1             = _mm256_add_pd(fix1,tx);
 482             fiy1             = _mm256_add_pd(fiy1,ty);
 483             fiz1             = _mm256_add_pd(fiz1,tz);
 484
 485             fjx0             = _mm256_add_pd(fjx0,tx);
 486             fjy0             = _mm256_add_pd(fjy0,ty);
 487             fjz0             = _mm256_add_pd(fjz0,tz);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r20              = _mm256_mul_pd(rsq20,rinv20);
 494             r20              = _mm256_andnot_pd(dummy_mask,r20);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq20             = _mm256_mul_pd(iq2,jq0);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 502             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 503             ewitab           = _mm256_cvttpd_epi32(ewrt);
 504             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 505             ewitab           = _mm_slli_epi32(ewitab,2);
 506             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 507             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 508             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 509             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 510             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 511             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 512             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 513             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 514             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm256_andnot_pd(dummy_mask,velec);
 518             velecsum         = _mm256_add_pd(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_pd(fscal,dx20);
 526             ty               = _mm256_mul_pd(fscal,dy20);
 527             tz               = _mm256_mul_pd(fscal,dz20);
 528
 529             /* Update vectorial force */
 530             fix2             = _mm256_add_pd(fix2,tx);
 531             fiy2             = _mm256_add_pd(fiy2,ty);
 532             fiz2             = _mm256_add_pd(fiz2,tz);
 533
 534             fjx0             = _mm256_add_pd(fjx0,tx);
 535             fjy0             = _mm256_add_pd(fjy0,ty);
 536             fjz0             = _mm256_add_pd(fjz0,tz);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r30              = _mm256_mul_pd(rsq30,rinv30);
 543             r30              = _mm256_andnot_pd(dummy_mask,r30);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq30             = _mm256_mul_pd(iq3,jq0);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 551             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 552             ewitab           = _mm256_cvttpd_epi32(ewrt);
 553             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 554             ewitab           = _mm_slli_epi32(ewitab,2);
 555             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 556             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 557             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 558             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 559             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 560             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 561             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 562             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 563             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm256_andnot_pd(dummy_mask,velec);
 567             velecsum         = _mm256_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx30);
 575             ty               = _mm256_mul_pd(fscal,dy30);
 576             tz               = _mm256_mul_pd(fscal,dz30);
 577
 578             /* Update vectorial force */
 579             fix3             = _mm256_add_pd(fix3,tx);
 580             fiy3             = _mm256_add_pd(fiy3,ty);
 581             fiz3             = _mm256_add_pd(fiz3,tz);
 582
 583             fjx0             = _mm256_add_pd(fjx0,tx);
 584             fjy0             = _mm256_add_pd(fjy0,ty);
 585             fjz0             = _mm256_add_pd(fjz0,tz);
 586
 587             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 588             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 589             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 590             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 591
 592             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 593
 594             /* Inner loop uses 129 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 600                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 601
 602         ggid                        = gid[iidx];
 603         /* Update potential energies */
 604         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 605
 606         /* Increment number of inner iterations */
 607         inneriter                  += j_index_end - j_index_start;
 608
 609         /* Outer loop uses 19 flops */
 610     }
 611
 612     /* Increment number of outer iterations */
 613     outeriter        += nri;
 614
 615     /* Update outer/inner flops */
 616
 617     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*129);
 618 }
 619 /*
 620  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_double
 621  * Electrostatics interaction: Ewald
 622  * VdW interaction:            None
 623  * Geometry:                   Water4-Particle
 624  * Calculate force/pot:        Force
 625  */
 626 void
 627 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_double
 628                     (t_nblist                    * gmx_restrict       nlist,
 629                      rvec                        * gmx_restrict          xx,
 630                      rvec                        * gmx_restrict          ff,
 631                      struct t_forcerec           * gmx_restrict          fr,
 632                      t_mdatoms                   * gmx_restrict     mdatoms,
 633                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 634                      t_nrnb                      * gmx_restrict        nrnb)
 635 {
 636     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 637      * just 0 for non-waters.
 638      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 639      * jnr indices corresponding to data put in the four positions in the SIMD register.
 640      */
 641     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 642     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 643     int              jnrA,jnrB,jnrC,jnrD;
 644     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 645     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 646     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 647     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 648     real             rcutoff_scalar;
 649     real             *shiftvec,*fshift,*x,*f;
 650     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 651     real             scratch[4*DIM];
 652     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 653     real *           vdwioffsetptr1;
 654     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 655     real *           vdwioffsetptr2;
 656     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 657     real *           vdwioffsetptr3;
 658     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 659     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 660     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 661     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 662     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 663     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 664     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 665     real             *charge;
 666     __m128i          ewitab;
 667     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 668     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 669     real             *ewtab;
 670     __m256d          dummy_mask,cutoff_mask;
 671     __m128           tmpmask0,tmpmask1;
 672     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 673     __m256d          one     = _mm256_set1_pd(1.0);
 674     __m256d          two     = _mm256_set1_pd(2.0);
 675     x                = xx[0];
 676     f                = ff[0];
 677
 678     nri              = nlist->nri;
 679     iinr             = nlist->iinr;
 680     jindex           = nlist->jindex;
 681     jjnr             = nlist->jjnr;
 682     shiftidx         = nlist->shift;
 683     gid              = nlist->gid;
 684     shiftvec         = fr->shift_vec[0];
 685     fshift           = fr->fshift[0];
 686     facel            = _mm256_set1_pd(fr->ic->epsfac);
 687     charge           = mdatoms->chargeA;
 688
 689     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 690     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 691     beta2            = _mm256_mul_pd(beta,beta);
 692     beta3            = _mm256_mul_pd(beta,beta2);
 693
 694     ewtab            = fr->ic->tabq_coul_F;
 695     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 696     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 697
 698     /* Setup water-specific parameters */
 699     inr              = nlist->iinr[0];
 700     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 701     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 702     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 703
 704     /* Avoid stupid compiler warnings */
 705     jnrA = jnrB = jnrC = jnrD = 0;
 706     j_coord_offsetA = 0;
 707     j_coord_offsetB = 0;
 708     j_coord_offsetC = 0;
 709     j_coord_offsetD = 0;
 710
 711     outeriter        = 0;
 712     inneriter        = 0;
 713
 714     for(iidx=0;iidx<4*DIM;iidx++)
 715     {
 716         scratch[iidx] = 0.0;
 717     }
 718
 719     /* Start outer loop over neighborlists */
 720     for(iidx=0; iidx<nri; iidx++)
 721     {
 722         /* Load shift vector for this list */
 723         i_shift_offset   = DIM*shiftidx[iidx];
 724
 725         /* Load limits for loop over neighbors */
 726         j_index_start    = jindex[iidx];
 727         j_index_end      = jindex[iidx+1];
 728
 729         /* Get outer coordinate index */
 730         inr              = iinr[iidx];
 731         i_coord_offset   = DIM*inr;
 732
 733         /* Load i particle coords and add shift vector */
 734         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 735                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 736
 737         fix1             = _mm256_setzero_pd();
 738         fiy1             = _mm256_setzero_pd();
 739         fiz1             = _mm256_setzero_pd();
 740         fix2             = _mm256_setzero_pd();
 741         fiy2             = _mm256_setzero_pd();
 742         fiz2             = _mm256_setzero_pd();
 743         fix3             = _mm256_setzero_pd();
 744         fiy3             = _mm256_setzero_pd();
 745         fiz3             = _mm256_setzero_pd();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             jnrC             = jjnr[jidx+2];
 755             jnrD             = jjnr[jidx+3];
 756             j_coord_offsetA  = DIM*jnrA;
 757             j_coord_offsetB  = DIM*jnrB;
 758             j_coord_offsetC  = DIM*jnrC;
 759             j_coord_offsetD  = DIM*jnrD;
 760
 761             /* load j atom coordinates */
 762             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 763                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 764                                                  &jx0,&jy0,&jz0);
 765
 766             /* Calculate displacement vector */
 767             dx10             = _mm256_sub_pd(ix1,jx0);
 768             dy10             = _mm256_sub_pd(iy1,jy0);
 769             dz10             = _mm256_sub_pd(iz1,jz0);
 770             dx20             = _mm256_sub_pd(ix2,jx0);
 771             dy20             = _mm256_sub_pd(iy2,jy0);
 772             dz20             = _mm256_sub_pd(iz2,jz0);
 773             dx30             = _mm256_sub_pd(ix3,jx0);
 774             dy30             = _mm256_sub_pd(iy3,jy0);
 775             dz30             = _mm256_sub_pd(iz3,jz0);
 776
 777             /* Calculate squared distance and things based on it */
 778             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 779             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 780             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 781
 782             rinv10           = avx256_invsqrt_d(rsq10);
 783             rinv20           = avx256_invsqrt_d(rsq20);
 784             rinv30           = avx256_invsqrt_d(rsq30);
 785
 786             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 787             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 788             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 789
 790             /* Load parameters for j particles */
 791             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 792                                                                  charge+jnrC+0,charge+jnrD+0);
 793
 794             fjx0             = _mm256_setzero_pd();
 795             fjy0             = _mm256_setzero_pd();
 796             fjz0             = _mm256_setzero_pd();
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             r10              = _mm256_mul_pd(rsq10,rinv10);
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq10             = _mm256_mul_pd(iq1,jq0);
 806
 807             /* EWALD ELECTROSTATICS */
 808
 809             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 810             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 811             ewitab           = _mm256_cvttpd_epi32(ewrt);
 812             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 813             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 814                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 815                                             &ewtabF,&ewtabFn);
 816             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 817             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 818
 819             fscal            = felec;
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm256_mul_pd(fscal,dx10);
 823             ty               = _mm256_mul_pd(fscal,dy10);
 824             tz               = _mm256_mul_pd(fscal,dz10);
 825
 826             /* Update vectorial force */
 827             fix1             = _mm256_add_pd(fix1,tx);
 828             fiy1             = _mm256_add_pd(fiy1,ty);
 829             fiz1             = _mm256_add_pd(fiz1,tz);
 830
 831             fjx0             = _mm256_add_pd(fjx0,tx);
 832             fjy0             = _mm256_add_pd(fjy0,ty);
 833             fjz0             = _mm256_add_pd(fjz0,tz);
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             r20              = _mm256_mul_pd(rsq20,rinv20);
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq20             = _mm256_mul_pd(iq2,jq0);
 843
 844             /* EWALD ELECTROSTATICS */
 845
 846             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 847             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 848             ewitab           = _mm256_cvttpd_epi32(ewrt);
 849             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 850             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 851                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 852                                             &ewtabF,&ewtabFn);
 853             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 854             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 855
 856             fscal            = felec;
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm256_mul_pd(fscal,dx20);
 860             ty               = _mm256_mul_pd(fscal,dy20);
 861             tz               = _mm256_mul_pd(fscal,dz20);
 862
 863             /* Update vectorial force */
 864             fix2             = _mm256_add_pd(fix2,tx);
 865             fiy2             = _mm256_add_pd(fiy2,ty);
 866             fiz2             = _mm256_add_pd(fiz2,tz);
 867
 868             fjx0             = _mm256_add_pd(fjx0,tx);
 869             fjy0             = _mm256_add_pd(fjy0,ty);
 870             fjz0             = _mm256_add_pd(fjz0,tz);
 871
 872             /**************************
 873              * CALCULATE INTERACTIONS *
 874              **************************/
 875
 876             r30              = _mm256_mul_pd(rsq30,rinv30);
 877
 878             /* Compute parameters for interactions between i and j atoms */
 879             qq30             = _mm256_mul_pd(iq3,jq0);
 880
 881             /* EWALD ELECTROSTATICS */
 882
 883             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 884             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 885             ewitab           = _mm256_cvttpd_epi32(ewrt);
 886             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 887             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 888                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 889                                             &ewtabF,&ewtabFn);
 890             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 891             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 892
 893             fscal            = felec;
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx30);
 897             ty               = _mm256_mul_pd(fscal,dy30);
 898             tz               = _mm256_mul_pd(fscal,dz30);
 899
 900             /* Update vectorial force */
 901             fix3             = _mm256_add_pd(fix3,tx);
 902             fiy3             = _mm256_add_pd(fiy3,ty);
 903             fiz3             = _mm256_add_pd(fiz3,tz);
 904
 905             fjx0             = _mm256_add_pd(fjx0,tx);
 906             fjy0             = _mm256_add_pd(fjy0,ty);
 907             fjz0             = _mm256_add_pd(fjz0,tz);
 908
 909             fjptrA             = f+j_coord_offsetA;
 910             fjptrB             = f+j_coord_offsetB;
 911             fjptrC             = f+j_coord_offsetC;
 912             fjptrD             = f+j_coord_offsetD;
 913
 914             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 915
 916             /* Inner loop uses 111 flops */
 917         }
 918
 919         if(jidx<j_index_end)
 920         {
 921
 922             /* Get j neighbor index, and coordinate index */
 923             jnrlistA         = jjnr[jidx];
 924             jnrlistB         = jjnr[jidx+1];
 925             jnrlistC         = jjnr[jidx+2];
 926             jnrlistD         = jjnr[jidx+3];
 927             /* Sign of each element will be negative for non-real atoms.
 928              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 929              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 930              */
 931             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 932
 933             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 934             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 935             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 936
 937             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 938             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 939             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 940             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 941             j_coord_offsetA  = DIM*jnrA;
 942             j_coord_offsetB  = DIM*jnrB;
 943             j_coord_offsetC  = DIM*jnrC;
 944             j_coord_offsetD  = DIM*jnrD;
 945
 946             /* load j atom coordinates */
 947             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 948                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 949                                                  &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx10             = _mm256_sub_pd(ix1,jx0);
 953             dy10             = _mm256_sub_pd(iy1,jy0);
 954             dz10             = _mm256_sub_pd(iz1,jz0);
 955             dx20             = _mm256_sub_pd(ix2,jx0);
 956             dy20             = _mm256_sub_pd(iy2,jy0);
 957             dz20             = _mm256_sub_pd(iz2,jz0);
 958             dx30             = _mm256_sub_pd(ix3,jx0);
 959             dy30             = _mm256_sub_pd(iy3,jy0);
 960             dz30             = _mm256_sub_pd(iz3,jz0);
 961
 962             /* Calculate squared distance and things based on it */
 963             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 964             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 965             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 966
 967             rinv10           = avx256_invsqrt_d(rsq10);
 968             rinv20           = avx256_invsqrt_d(rsq20);
 969             rinv30           = avx256_invsqrt_d(rsq30);
 970
 971             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 972             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 973             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 974
 975             /* Load parameters for j particles */
 976             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 977                                                                  charge+jnrC+0,charge+jnrD+0);
 978
 979             fjx0             = _mm256_setzero_pd();
 980             fjy0             = _mm256_setzero_pd();
 981             fjz0             = _mm256_setzero_pd();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r10              = _mm256_mul_pd(rsq10,rinv10);
 988             r10              = _mm256_andnot_pd(dummy_mask,r10);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm256_mul_pd(iq1,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 996             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 997             ewitab           = _mm256_cvttpd_epi32(ewrt);
 998             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 999             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1000                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1001                                             &ewtabF,&ewtabFn);
1002             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1003             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1004
1005             fscal            = felec;
1006
1007             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm256_mul_pd(fscal,dx10);
1011             ty               = _mm256_mul_pd(fscal,dy10);
1012             tz               = _mm256_mul_pd(fscal,dz10);
1013
1014             /* Update vectorial force */
1015             fix1             = _mm256_add_pd(fix1,tx);
1016             fiy1             = _mm256_add_pd(fiy1,ty);
1017             fiz1             = _mm256_add_pd(fiz1,tz);
1018
1019             fjx0             = _mm256_add_pd(fjx0,tx);
1020             fjy0             = _mm256_add_pd(fjy0,ty);
1021             fjz0             = _mm256_add_pd(fjz0,tz);
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             r20              = _mm256_mul_pd(rsq20,rinv20);
1028             r20              = _mm256_andnot_pd(dummy_mask,r20);
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq20             = _mm256_mul_pd(iq2,jq0);
1032
1033             /* EWALD ELECTROSTATICS */
1034
1035             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1036             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1037             ewitab           = _mm256_cvttpd_epi32(ewrt);
1038             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1039             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1040                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1041                                             &ewtabF,&ewtabFn);
1042             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1043             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm256_mul_pd(fscal,dx20);
1051             ty               = _mm256_mul_pd(fscal,dy20);
1052             tz               = _mm256_mul_pd(fscal,dz20);
1053
1054             /* Update vectorial force */
1055             fix2             = _mm256_add_pd(fix2,tx);
1056             fiy2             = _mm256_add_pd(fiy2,ty);
1057             fiz2             = _mm256_add_pd(fiz2,tz);
1058
1059             fjx0             = _mm256_add_pd(fjx0,tx);
1060             fjy0             = _mm256_add_pd(fjy0,ty);
1061             fjz0             = _mm256_add_pd(fjz0,tz);
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             r30              = _mm256_mul_pd(rsq30,rinv30);
1068             r30              = _mm256_andnot_pd(dummy_mask,r30);
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq30             = _mm256_mul_pd(iq3,jq0);
1072
1073             /* EWALD ELECTROSTATICS */
1074
1075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1076             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1077             ewitab           = _mm256_cvttpd_epi32(ewrt);
1078             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1079             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1080                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1081                                             &ewtabF,&ewtabFn);
1082             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1083             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_pd(fscal,dx30);
1091             ty               = _mm256_mul_pd(fscal,dy30);
1092             tz               = _mm256_mul_pd(fscal,dz30);
1093
1094             /* Update vectorial force */
1095             fix3             = _mm256_add_pd(fix3,tx);
1096             fiy3             = _mm256_add_pd(fiy3,ty);
1097             fiz3             = _mm256_add_pd(fiz3,tz);
1098
1099             fjx0             = _mm256_add_pd(fjx0,tx);
1100             fjy0             = _mm256_add_pd(fjy0,ty);
1101             fjz0             = _mm256_add_pd(fjz0,tz);
1102
1103             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1104             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1105             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1106             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1107
1108             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1109
1110             /* Inner loop uses 114 flops */
1111         }
1112
1113         /* End of innermost loop */
1114
1115         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1116                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1117
1118         /* Increment number of inner iterations */
1119         inneriter                  += j_index_end - j_index_start;
1120
1121         /* Outer loop uses 18 flops */
1122     }
1123
1124     /* Increment number of outer iterations */
1125     outeriter        += nri;
1126
1127     /* Update outer/inner flops */
1128
1129     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*114);
1130 }