src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEwSh_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_ElecEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm256_setzero_pd();
 176         fiy0             = _mm256_setzero_pd();
 177         fiz0             = _mm256_setzero_pd();
 178         fix1             = _mm256_setzero_pd();
 179         fiy1             = _mm256_setzero_pd();
 180         fiz1             = _mm256_setzero_pd();
 181         fix2             = _mm256_setzero_pd();
 182         fiy2             = _mm256_setzero_pd();
 183         fiz2             = _mm256_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_pd(ix0,jx0);
 209             dy00             = _mm256_sub_pd(iy0,jy0);
 210             dz00             = _mm256_sub_pd(iz0,jz0);
 211             dx10             = _mm256_sub_pd(ix1,jx0);
 212             dy10             = _mm256_sub_pd(iy1,jy0);
 213             dz10             = _mm256_sub_pd(iz1,jz0);
 214             dx20             = _mm256_sub_pd(ix2,jx0);
 215             dy20             = _mm256_sub_pd(iy2,jy0);
 216             dz20             = _mm256_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 226
 227             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 228             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0);
 234
 235             fjx0             = _mm256_setzero_pd();
 236             fjy0             = _mm256_setzero_pd();
 237             fjz0             = _mm256_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 244             {
 245
 246             r00              = _mm256_mul_pd(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm256_mul_pd(iq0,jq0);
 250
 251             /* EWALD ELECTROSTATICS */
 252
 253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 254             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 255             ewitab           = _mm256_cvttpd_epi32(ewrt);
 256             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 257             ewitab           = _mm_slli_epi32(ewitab,2);
 258             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 259             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 260             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 261             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 262             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 263             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 264             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 265             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 266             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 267
 268             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velec            = _mm256_and_pd(velec,cutoff_mask);
 272             velecsum         = _mm256_add_pd(velecsum,velec);
 273
 274             fscal            = felec;
 275
 276             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm256_mul_pd(fscal,dx00);
 280             ty               = _mm256_mul_pd(fscal,dy00);
 281             tz               = _mm256_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm256_add_pd(fix0,tx);
 285             fiy0             = _mm256_add_pd(fiy0,ty);
 286             fiz0             = _mm256_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm256_add_pd(fjx0,tx);
 289             fjy0             = _mm256_add_pd(fjy0,ty);
 290             fjz0             = _mm256_add_pd(fjz0,tz);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 299             {
 300
 301             r10              = _mm256_mul_pd(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm256_mul_pd(iq1,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 310             ewitab           = _mm256_cvttpd_epi32(ewrt);
 311             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 315             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 316             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 317             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 318             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 319             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 320             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 321             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 322
 323             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm256_and_pd(velec,cutoff_mask);
 327             velecsum         = _mm256_add_pd(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_pd(fscal,dx10);
 335             ty               = _mm256_mul_pd(fscal,dy10);
 336             tz               = _mm256_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm256_add_pd(fix1,tx);
 340             fiy1             = _mm256_add_pd(fiy1,ty);
 341             fiz1             = _mm256_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm256_add_pd(fjx0,tx);
 344             fjy0             = _mm256_add_pd(fjy0,ty);
 345             fjz0             = _mm256_add_pd(fjz0,tz);
 346
 347             }
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 354             {
 355
 356             r20              = _mm256_mul_pd(rsq20,rinv20);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm256_mul_pd(iq2,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 365             ewitab           = _mm256_cvttpd_epi32(ewrt);
 366             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 367             ewitab           = _mm_slli_epi32(ewitab,2);
 368             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 369             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 370             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 371             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 372             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 373             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 374             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 375             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 376             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 377
 378             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm256_and_pd(velec,cutoff_mask);
 382             velecsum         = _mm256_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm256_mul_pd(fscal,dx20);
 390             ty               = _mm256_mul_pd(fscal,dy20);
 391             tz               = _mm256_mul_pd(fscal,dz20);
 392
 393             /* Update vectorial force */
 394             fix2             = _mm256_add_pd(fix2,tx);
 395             fiy2             = _mm256_add_pd(fiy2,ty);
 396             fiz2             = _mm256_add_pd(fiz2,tz);
 397
 398             fjx0             = _mm256_add_pd(fjx0,tx);
 399             fjy0             = _mm256_add_pd(fjy0,ty);
 400             fjz0             = _mm256_add_pd(fjz0,tz);
 401
 402             }
 403
 404             fjptrA             = f+j_coord_offsetA;
 405             fjptrB             = f+j_coord_offsetB;
 406             fjptrC             = f+j_coord_offsetC;
 407             fjptrD             = f+j_coord_offsetD;
 408
 409             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 141 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrlistA         = jjnr[jidx];
 419             jnrlistB         = jjnr[jidx+1];
 420             jnrlistC         = jjnr[jidx+2];
 421             jnrlistD         = jjnr[jidx+3];
 422             /* Sign of each element will be negative for non-real atoms.
 423              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 424              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 425              */
 426             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 427
 428             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 429             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 430             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 431
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 444                                                  &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm256_sub_pd(ix0,jx0);
 448             dy00             = _mm256_sub_pd(iy0,jy0);
 449             dz00             = _mm256_sub_pd(iz0,jz0);
 450             dx10             = _mm256_sub_pd(ix1,jx0);
 451             dy10             = _mm256_sub_pd(iy1,jy0);
 452             dz10             = _mm256_sub_pd(iz1,jz0);
 453             dx20             = _mm256_sub_pd(ix2,jx0);
 454             dy20             = _mm256_sub_pd(iy2,jy0);
 455             dz20             = _mm256_sub_pd(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 459             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 460             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 463             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 464             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 465
 466             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 467             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 468             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 472                                                                  charge+jnrC+0,charge+jnrD+0);
 473
 474             fjx0             = _mm256_setzero_pd();
 475             fjy0             = _mm256_setzero_pd();
 476             fjz0             = _mm256_setzero_pd();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 483             {
 484
 485             r00              = _mm256_mul_pd(rsq00,rinv00);
 486             r00              = _mm256_andnot_pd(dummy_mask,r00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm256_mul_pd(iq0,jq0);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 495             ewitab           = _mm256_cvttpd_epi32(ewrt);
 496             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 497             ewitab           = _mm_slli_epi32(ewitab,2);
 498             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 499             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 500             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 501             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 502             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 503             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 504             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 505             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 506             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 507
 508             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm256_and_pd(velec,cutoff_mask);
 512             velec            = _mm256_andnot_pd(dummy_mask,velec);
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx00);
 523             ty               = _mm256_mul_pd(fscal,dy00);
 524             tz               = _mm256_mul_pd(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm256_add_pd(fix0,tx);
 528             fiy0             = _mm256_add_pd(fiy0,ty);
 529             fiz0             = _mm256_add_pd(fiz0,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             }
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 542             {
 543
 544             r10              = _mm256_mul_pd(rsq10,rinv10);
 545             r10              = _mm256_andnot_pd(dummy_mask,r10);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq10             = _mm256_mul_pd(iq1,jq0);
 549
 550             /* EWALD ELECTROSTATICS */
 551
 552             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 553             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 554             ewitab           = _mm256_cvttpd_epi32(ewrt);
 555             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 556             ewitab           = _mm_slli_epi32(ewitab,2);
 557             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 558             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 559             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 560             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 561             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 562             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 563             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 564             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 565             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 566
 567             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm256_and_pd(velec,cutoff_mask);
 571             velec            = _mm256_andnot_pd(dummy_mask,velec);
 572             velecsum         = _mm256_add_pd(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 577
 578             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm256_mul_pd(fscal,dx10);
 582             ty               = _mm256_mul_pd(fscal,dy10);
 583             tz               = _mm256_mul_pd(fscal,dz10);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm256_add_pd(fix1,tx);
 587             fiy1             = _mm256_add_pd(fiy1,ty);
 588             fiz1             = _mm256_add_pd(fiz1,tz);
 589
 590             fjx0             = _mm256_add_pd(fjx0,tx);
 591             fjy0             = _mm256_add_pd(fjy0,ty);
 592             fjz0             = _mm256_add_pd(fjz0,tz);
 593
 594             }
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 601             {
 602
 603             r20              = _mm256_mul_pd(rsq20,rinv20);
 604             r20              = _mm256_andnot_pd(dummy_mask,r20);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq20             = _mm256_mul_pd(iq2,jq0);
 608
 609             /* EWALD ELECTROSTATICS */
 610
 611             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 612             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 613             ewitab           = _mm256_cvttpd_epi32(ewrt);
 614             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 615             ewitab           = _mm_slli_epi32(ewitab,2);
 616             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 617             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 618             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 619             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 620             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 621             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 622             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 623             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 624             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 625
 626             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 627
 628             /* Update potential sum for this i atom from the interaction with this j atom. */
 629             velec            = _mm256_and_pd(velec,cutoff_mask);
 630             velec            = _mm256_andnot_pd(dummy_mask,velec);
 631             velecsum         = _mm256_add_pd(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 636
 637             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_pd(fscal,dx20);
 641             ty               = _mm256_mul_pd(fscal,dy20);
 642             tz               = _mm256_mul_pd(fscal,dz20);
 643
 644             /* Update vectorial force */
 645             fix2             = _mm256_add_pd(fix2,tx);
 646             fiy2             = _mm256_add_pd(fiy2,ty);
 647             fiz2             = _mm256_add_pd(fiz2,tz);
 648
 649             fjx0             = _mm256_add_pd(fjx0,tx);
 650             fjy0             = _mm256_add_pd(fjy0,ty);
 651             fjz0             = _mm256_add_pd(fjz0,tz);
 652
 653             }
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659
 660             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 661
 662             /* Inner loop uses 144 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 668                                                  f+i_coord_offset,fshift+i_shift_offset);
 669
 670         ggid                        = gid[iidx];
 671         /* Update potential energies */
 672         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 673
 674         /* Increment number of inner iterations */
 675         inneriter                  += j_index_end - j_index_start;
 676
 677         /* Outer loop uses 19 flops */
 678     }
 679
 680     /* Increment number of outer iterations */
 681     outeriter        += nri;
 682
 683     /* Update outer/inner flops */
 684
 685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*144);
 686 }
 687 /*
 688  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_double
 689  * Electrostatics interaction: Ewald
 690  * VdW interaction:            None
 691  * Geometry:                   Water3-Particle
 692  * Calculate force/pot:        Force
 693  */
 694 void
 695 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_double
 696                     (t_nblist                    * gmx_restrict       nlist,
 697                      rvec                        * gmx_restrict          xx,
 698                      rvec                        * gmx_restrict          ff,
 699                      t_forcerec                  * gmx_restrict          fr,
 700                      t_mdatoms                   * gmx_restrict     mdatoms,
 701                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 702                      t_nrnb                      * gmx_restrict        nrnb)
 703 {
 704     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 705      * just 0 for non-waters.
 706      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 707      * jnr indices corresponding to data put in the four positions in the SIMD register.
 708      */
 709     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 710     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 711     int              jnrA,jnrB,jnrC,jnrD;
 712     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 713     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 714     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 715     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 716     real             rcutoff_scalar;
 717     real             *shiftvec,*fshift,*x,*f;
 718     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 719     real             scratch[4*DIM];
 720     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 721     real *           vdwioffsetptr0;
 722     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 723     real *           vdwioffsetptr1;
 724     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 725     real *           vdwioffsetptr2;
 726     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 727     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 728     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 729     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 730     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 731     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 732     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 733     real             *charge;
 734     __m128i          ewitab;
 735     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 736     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 737     real             *ewtab;
 738     __m256d          dummy_mask,cutoff_mask;
 739     __m128           tmpmask0,tmpmask1;
 740     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 741     __m256d          one     = _mm256_set1_pd(1.0);
 742     __m256d          two     = _mm256_set1_pd(2.0);
 743     x                = xx[0];
 744     f                = ff[0];
 745
 746     nri              = nlist->nri;
 747     iinr             = nlist->iinr;
 748     jindex           = nlist->jindex;
 749     jjnr             = nlist->jjnr;
 750     shiftidx         = nlist->shift;
 751     gid              = nlist->gid;
 752     shiftvec         = fr->shift_vec[0];
 753     fshift           = fr->fshift[0];
 754     facel            = _mm256_set1_pd(fr->epsfac);
 755     charge           = mdatoms->chargeA;
 756
 757     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 758     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 759     beta2            = _mm256_mul_pd(beta,beta);
 760     beta3            = _mm256_mul_pd(beta,beta2);
 761
 762     ewtab            = fr->ic->tabq_coul_F;
 763     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 764     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 765
 766     /* Setup water-specific parameters */
 767     inr              = nlist->iinr[0];
 768     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 769     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 770     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 771
 772     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 773     rcutoff_scalar   = fr->rcoulomb;
 774     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 775     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 776
 777     /* Avoid stupid compiler warnings */
 778     jnrA = jnrB = jnrC = jnrD = 0;
 779     j_coord_offsetA = 0;
 780     j_coord_offsetB = 0;
 781     j_coord_offsetC = 0;
 782     j_coord_offsetD = 0;
 783
 784     outeriter        = 0;
 785     inneriter        = 0;
 786
 787     for(iidx=0;iidx<4*DIM;iidx++)
 788     {
 789         scratch[iidx] = 0.0;
 790     }
 791
 792     /* Start outer loop over neighborlists */
 793     for(iidx=0; iidx<nri; iidx++)
 794     {
 795         /* Load shift vector for this list */
 796         i_shift_offset   = DIM*shiftidx[iidx];
 797
 798         /* Load limits for loop over neighbors */
 799         j_index_start    = jindex[iidx];
 800         j_index_end      = jindex[iidx+1];
 801
 802         /* Get outer coordinate index */
 803         inr              = iinr[iidx];
 804         i_coord_offset   = DIM*inr;
 805
 806         /* Load i particle coords and add shift vector */
 807         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 808                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 809
 810         fix0             = _mm256_setzero_pd();
 811         fiy0             = _mm256_setzero_pd();
 812         fiz0             = _mm256_setzero_pd();
 813         fix1             = _mm256_setzero_pd();
 814         fiy1             = _mm256_setzero_pd();
 815         fiz1             = _mm256_setzero_pd();
 816         fix2             = _mm256_setzero_pd();
 817         fiy2             = _mm256_setzero_pd();
 818         fiz2             = _mm256_setzero_pd();
 819
 820         /* Start inner kernel loop */
 821         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 822         {
 823
 824             /* Get j neighbor index, and coordinate index */
 825             jnrA             = jjnr[jidx];
 826             jnrB             = jjnr[jidx+1];
 827             jnrC             = jjnr[jidx+2];
 828             jnrD             = jjnr[jidx+3];
 829             j_coord_offsetA  = DIM*jnrA;
 830             j_coord_offsetB  = DIM*jnrB;
 831             j_coord_offsetC  = DIM*jnrC;
 832             j_coord_offsetD  = DIM*jnrD;
 833
 834             /* load j atom coordinates */
 835             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 836                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 837                                                  &jx0,&jy0,&jz0);
 838
 839             /* Calculate displacement vector */
 840             dx00             = _mm256_sub_pd(ix0,jx0);
 841             dy00             = _mm256_sub_pd(iy0,jy0);
 842             dz00             = _mm256_sub_pd(iz0,jz0);
 843             dx10             = _mm256_sub_pd(ix1,jx0);
 844             dy10             = _mm256_sub_pd(iy1,jy0);
 845             dz10             = _mm256_sub_pd(iz1,jz0);
 846             dx20             = _mm256_sub_pd(ix2,jx0);
 847             dy20             = _mm256_sub_pd(iy2,jy0);
 848             dz20             = _mm256_sub_pd(iz2,jz0);
 849
 850             /* Calculate squared distance and things based on it */
 851             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 852             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 853             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 854
 855             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 856             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 857             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 858
 859             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 860             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 861             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 862
 863             /* Load parameters for j particles */
 864             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 865                                                                  charge+jnrC+0,charge+jnrD+0);
 866
 867             fjx0             = _mm256_setzero_pd();
 868             fjy0             = _mm256_setzero_pd();
 869             fjz0             = _mm256_setzero_pd();
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 876             {
 877
 878             r00              = _mm256_mul_pd(rsq00,rinv00);
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq00             = _mm256_mul_pd(iq0,jq0);
 882
 883             /* EWALD ELECTROSTATICS */
 884
 885             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 886             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 887             ewitab           = _mm256_cvttpd_epi32(ewrt);
 888             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 889             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 890                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 891                                             &ewtabF,&ewtabFn);
 892             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 893             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 894
 895             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 896
 897             fscal            = felec;
 898
 899             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm256_mul_pd(fscal,dx00);
 903             ty               = _mm256_mul_pd(fscal,dy00);
 904             tz               = _mm256_mul_pd(fscal,dz00);
 905
 906             /* Update vectorial force */
 907             fix0             = _mm256_add_pd(fix0,tx);
 908             fiy0             = _mm256_add_pd(fiy0,ty);
 909             fiz0             = _mm256_add_pd(fiz0,tz);
 910
 911             fjx0             = _mm256_add_pd(fjx0,tx);
 912             fjy0             = _mm256_add_pd(fjy0,ty);
 913             fjz0             = _mm256_add_pd(fjz0,tz);
 914
 915             }
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 922             {
 923
 924             r10              = _mm256_mul_pd(rsq10,rinv10);
 925
 926             /* Compute parameters for interactions between i and j atoms */
 927             qq10             = _mm256_mul_pd(iq1,jq0);
 928
 929             /* EWALD ELECTROSTATICS */
 930
 931             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 932             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 933             ewitab           = _mm256_cvttpd_epi32(ewrt);
 934             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 935             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 936                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 937                                             &ewtabF,&ewtabFn);
 938             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 939             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 940
 941             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 942
 943             fscal            = felec;
 944
 945             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 946
 947             /* Calculate temporary vectorial force */
 948             tx               = _mm256_mul_pd(fscal,dx10);
 949             ty               = _mm256_mul_pd(fscal,dy10);
 950             tz               = _mm256_mul_pd(fscal,dz10);
 951
 952             /* Update vectorial force */
 953             fix1             = _mm256_add_pd(fix1,tx);
 954             fiy1             = _mm256_add_pd(fiy1,ty);
 955             fiz1             = _mm256_add_pd(fiz1,tz);
 956
 957             fjx0             = _mm256_add_pd(fjx0,tx);
 958             fjy0             = _mm256_add_pd(fjy0,ty);
 959             fjz0             = _mm256_add_pd(fjz0,tz);
 960
 961             }
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 968             {
 969
 970             r20              = _mm256_mul_pd(rsq20,rinv20);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq20             = _mm256_mul_pd(iq2,jq0);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 978             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 979             ewitab           = _mm256_cvttpd_epi32(ewrt);
 980             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 981             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 982                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 983                                             &ewtabF,&ewtabFn);
 984             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 985             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 986
 987             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx20);
 995             ty               = _mm256_mul_pd(fscal,dy20);
 996             tz               = _mm256_mul_pd(fscal,dz20);
 997
 998             /* Update vectorial force */
 999             fix2             = _mm256_add_pd(fix2,tx);
1000             fiy2             = _mm256_add_pd(fiy2,ty);
1001             fiz2             = _mm256_add_pd(fiz2,tz);
1002
1003             fjx0             = _mm256_add_pd(fjx0,tx);
1004             fjy0             = _mm256_add_pd(fjy0,ty);
1005             fjz0             = _mm256_add_pd(fjz0,tz);
1006
1007             }
1008
1009             fjptrA             = f+j_coord_offsetA;
1010             fjptrB             = f+j_coord_offsetB;
1011             fjptrC             = f+j_coord_offsetC;
1012             fjptrD             = f+j_coord_offsetD;
1013
1014             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1015
1016             /* Inner loop uses 120 flops */
1017         }
1018
1019         if(jidx<j_index_end)
1020         {
1021
1022             /* Get j neighbor index, and coordinate index */
1023             jnrlistA         = jjnr[jidx];
1024             jnrlistB         = jjnr[jidx+1];
1025             jnrlistC         = jjnr[jidx+2];
1026             jnrlistD         = jjnr[jidx+3];
1027             /* Sign of each element will be negative for non-real atoms.
1028              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1029              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1030              */
1031             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1032
1033             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1034             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1035             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1036
1037             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1038             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1039             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1040             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1041             j_coord_offsetA  = DIM*jnrA;
1042             j_coord_offsetB  = DIM*jnrB;
1043             j_coord_offsetC  = DIM*jnrC;
1044             j_coord_offsetD  = DIM*jnrD;
1045
1046             /* load j atom coordinates */
1047             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1048                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1049                                                  &jx0,&jy0,&jz0);
1050
1051             /* Calculate displacement vector */
1052             dx00             = _mm256_sub_pd(ix0,jx0);
1053             dy00             = _mm256_sub_pd(iy0,jy0);
1054             dz00             = _mm256_sub_pd(iz0,jz0);
1055             dx10             = _mm256_sub_pd(ix1,jx0);
1056             dy10             = _mm256_sub_pd(iy1,jy0);
1057             dz10             = _mm256_sub_pd(iz1,jz0);
1058             dx20             = _mm256_sub_pd(ix2,jx0);
1059             dy20             = _mm256_sub_pd(iy2,jy0);
1060             dz20             = _mm256_sub_pd(iz2,jz0);
1061
1062             /* Calculate squared distance and things based on it */
1063             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1064             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1065             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1066
1067             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1068             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1069             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1070
1071             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1072             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1073             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1074
1075             /* Load parameters for j particles */
1076             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1077                                                                  charge+jnrC+0,charge+jnrD+0);
1078
1079             fjx0             = _mm256_setzero_pd();
1080             fjy0             = _mm256_setzero_pd();
1081             fjz0             = _mm256_setzero_pd();
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1088             {
1089
1090             r00              = _mm256_mul_pd(rsq00,rinv00);
1091             r00              = _mm256_andnot_pd(dummy_mask,r00);
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq00             = _mm256_mul_pd(iq0,jq0);
1095
1096             /* EWALD ELECTROSTATICS */
1097
1098             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1099             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1100             ewitab           = _mm256_cvttpd_epi32(ewrt);
1101             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1102             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1103                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1104                                             &ewtabF,&ewtabFn);
1105             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1106             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1107
1108             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1109
1110             fscal            = felec;
1111
1112             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1113
1114             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1115
1116             /* Calculate temporary vectorial force */
1117             tx               = _mm256_mul_pd(fscal,dx00);
1118             ty               = _mm256_mul_pd(fscal,dy00);
1119             tz               = _mm256_mul_pd(fscal,dz00);
1120
1121             /* Update vectorial force */
1122             fix0             = _mm256_add_pd(fix0,tx);
1123             fiy0             = _mm256_add_pd(fiy0,ty);
1124             fiz0             = _mm256_add_pd(fiz0,tz);
1125
1126             fjx0             = _mm256_add_pd(fjx0,tx);
1127             fjy0             = _mm256_add_pd(fjy0,ty);
1128             fjz0             = _mm256_add_pd(fjz0,tz);
1129
1130             }
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1137             {
1138
1139             r10              = _mm256_mul_pd(rsq10,rinv10);
1140             r10              = _mm256_andnot_pd(dummy_mask,r10);
1141
1142             /* Compute parameters for interactions between i and j atoms */
1143             qq10             = _mm256_mul_pd(iq1,jq0);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1148             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1149             ewitab           = _mm256_cvttpd_epi32(ewrt);
1150             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1151             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1152                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1153                                             &ewtabF,&ewtabFn);
1154             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1155             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1156
1157             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1162
1163             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm256_mul_pd(fscal,dx10);
1167             ty               = _mm256_mul_pd(fscal,dy10);
1168             tz               = _mm256_mul_pd(fscal,dz10);
1169
1170             /* Update vectorial force */
1171             fix1             = _mm256_add_pd(fix1,tx);
1172             fiy1             = _mm256_add_pd(fiy1,ty);
1173             fiz1             = _mm256_add_pd(fiz1,tz);
1174
1175             fjx0             = _mm256_add_pd(fjx0,tx);
1176             fjy0             = _mm256_add_pd(fjy0,ty);
1177             fjz0             = _mm256_add_pd(fjz0,tz);
1178
1179             }
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1186             {
1187
1188             r20              = _mm256_mul_pd(rsq20,rinv20);
1189             r20              = _mm256_andnot_pd(dummy_mask,r20);
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq20             = _mm256_mul_pd(iq2,jq0);
1193
1194             /* EWALD ELECTROSTATICS */
1195
1196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1197             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1198             ewitab           = _mm256_cvttpd_epi32(ewrt);
1199             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1200             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1201                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1202                                             &ewtabF,&ewtabFn);
1203             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1204             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1205
1206             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1207
1208             fscal            = felec;
1209
1210             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1211
1212             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1213
1214             /* Calculate temporary vectorial force */
1215             tx               = _mm256_mul_pd(fscal,dx20);
1216             ty               = _mm256_mul_pd(fscal,dy20);
1217             tz               = _mm256_mul_pd(fscal,dz20);
1218
1219             /* Update vectorial force */
1220             fix2             = _mm256_add_pd(fix2,tx);
1221             fiy2             = _mm256_add_pd(fiy2,ty);
1222             fiz2             = _mm256_add_pd(fiz2,tz);
1223
1224             fjx0             = _mm256_add_pd(fjx0,tx);
1225             fjy0             = _mm256_add_pd(fjy0,ty);
1226             fjz0             = _mm256_add_pd(fjz0,tz);
1227
1228             }
1229
1230             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1231             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1232             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1233             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1234
1235             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1236
1237             /* Inner loop uses 123 flops */
1238         }
1239
1240         /* End of innermost loop */
1241
1242         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1243                                                  f+i_coord_offset,fshift+i_shift_offset);
1244
1245         /* Increment number of inner iterations */
1246         inneriter                  += j_index_end - j_index_start;
1247
1248         /* Outer loop uses 18 flops */
1249     }
1250
1251     /* Increment number of outer iterations */
1252     outeriter        += nri;
1253
1254     /* Update outer/inner flops */
1255
1256     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*123);
1257 }