src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwNone_GeomP1P1_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          ewitab;
  94     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  96     real             *ewtab;
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 117     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 118     beta2            = _mm256_mul_pd(beta,beta);
 119     beta3            = _mm256_mul_pd(beta,beta2);
 120
 121     ewtab            = fr->ic->tabq_coul_FDV0;
 122     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 123     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm256_setzero_pd();
 158         fiy0             = _mm256_setzero_pd();
 159         fiz0             = _mm256_setzero_pd();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 184                                                  &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm256_sub_pd(ix0,jx0);
 188             dy00             = _mm256_sub_pd(iy0,jy0);
 189             dz00             = _mm256_sub_pd(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 195
 196             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 200                                                                  charge+jnrC+0,charge+jnrD+0);
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             r00              = _mm256_mul_pd(rsq00,rinv00);
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm256_mul_pd(iq0,jq0);
 210
 211             /* EWALD ELECTROSTATICS */
 212
 213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 214             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 215             ewitab           = _mm256_cvttpd_epi32(ewrt);
 216             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 217             ewitab           = _mm_slli_epi32(ewitab,2);
 218             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 219             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 220             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 221             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 222             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 223             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 224             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 225             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 226             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _mm256_add_pd(velecsum,velec);
 230
 231             fscal            = felec;
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm256_mul_pd(fscal,dx00);
 235             ty               = _mm256_mul_pd(fscal,dy00);
 236             tz               = _mm256_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm256_add_pd(fix0,tx);
 240             fiy0             = _mm256_add_pd(fiy0,ty);
 241             fiz0             = _mm256_add_pd(fiz0,tz);
 242
 243             fjptrA             = f+j_coord_offsetA;
 244             fjptrB             = f+j_coord_offsetB;
 245             fjptrC             = f+j_coord_offsetC;
 246             fjptrD             = f+j_coord_offsetD;
 247             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 248
 249             /* Inner loop uses 41 flops */
 250         }
 251
 252         if(jidx<j_index_end)
 253         {
 254
 255             /* Get j neighbor index, and coordinate index */
 256             jnrlistA         = jjnr[jidx];
 257             jnrlistB         = jjnr[jidx+1];
 258             jnrlistC         = jjnr[jidx+2];
 259             jnrlistD         = jjnr[jidx+3];
 260             /* Sign of each element will be negative for non-real atoms.
 261              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 262              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 263              */
 264             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 265
 266             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 267             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 268             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 269
 270             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 271             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 272             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 273             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 282                                                  &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm256_sub_pd(ix0,jx0);
 286             dy00             = _mm256_sub_pd(iy0,jy0);
 287             dz00             = _mm256_sub_pd(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 291
 292             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 293
 294             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 295
 296             /* Load parameters for j particles */
 297             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 298                                                                  charge+jnrC+0,charge+jnrD+0);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r00              = _mm256_mul_pd(rsq00,rinv00);
 305             r00              = _mm256_andnot_pd(dummy_mask,r00);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq00             = _mm256_mul_pd(iq0,jq0);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 314             ewitab           = _mm256_cvttpd_epi32(ewrt);
 315             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 319             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 320             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 321             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 322             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 323             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 324             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 325             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velec            = _mm256_andnot_pd(dummy_mask,velec);
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx00);
 337             ty               = _mm256_mul_pd(fscal,dy00);
 338             tz               = _mm256_mul_pd(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_pd(fix0,tx);
 342             fiy0             = _mm256_add_pd(fiy0,ty);
 343             fiz0             = _mm256_add_pd(fiz0,tz);
 344
 345             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 346             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 347             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 348             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 349             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 350
 351             /* Inner loop uses 42 flops */
 352         }
 353
 354         /* End of innermost loop */
 355
 356         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 357                                                  f+i_coord_offset,fshift+i_shift_offset);
 358
 359         ggid                        = gid[iidx];
 360         /* Update potential energies */
 361         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 362
 363         /* Increment number of inner iterations */
 364         inneriter                  += j_index_end - j_index_start;
 365
 366         /* Outer loop uses 8 flops */
 367     }
 368
 369     /* Increment number of outer iterations */
 370     outeriter        += nri;
 371
 372     /* Update outer/inner flops */
 373
 374     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
 375 }
 376 /*
 377  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_double
 378  * Electrostatics interaction: Ewald
 379  * VdW interaction:            None
 380  * Geometry:                   Particle-Particle
 381  * Calculate force/pot:        Force
 382  */
 383 void
 384 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_double
 385                     (t_nblist                    * gmx_restrict       nlist,
 386                      rvec                        * gmx_restrict          xx,
 387                      rvec                        * gmx_restrict          ff,
 388                      t_forcerec                  * gmx_restrict          fr,
 389                      t_mdatoms                   * gmx_restrict     mdatoms,
 390                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 391                      t_nrnb                      * gmx_restrict        nrnb)
 392 {
 393     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 394      * just 0 for non-waters.
 395      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 396      * jnr indices corresponding to data put in the four positions in the SIMD register.
 397      */
 398     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 399     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 400     int              jnrA,jnrB,jnrC,jnrD;
 401     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 402     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 403     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 404     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 405     real             rcutoff_scalar;
 406     real             *shiftvec,*fshift,*x,*f;
 407     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 408     real             scratch[4*DIM];
 409     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 410     real *           vdwioffsetptr0;
 411     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 412     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 413     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 414     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 415     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 416     real             *charge;
 417     __m128i          ewitab;
 418     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 419     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 420     real             *ewtab;
 421     __m256d          dummy_mask,cutoff_mask;
 422     __m128           tmpmask0,tmpmask1;
 423     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 424     __m256d          one     = _mm256_set1_pd(1.0);
 425     __m256d          two     = _mm256_set1_pd(2.0);
 426     x                = xx[0];
 427     f                = ff[0];
 428
 429     nri              = nlist->nri;
 430     iinr             = nlist->iinr;
 431     jindex           = nlist->jindex;
 432     jjnr             = nlist->jjnr;
 433     shiftidx         = nlist->shift;
 434     gid              = nlist->gid;
 435     shiftvec         = fr->shift_vec[0];
 436     fshift           = fr->fshift[0];
 437     facel            = _mm256_set1_pd(fr->epsfac);
 438     charge           = mdatoms->chargeA;
 439
 440     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 441     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 442     beta2            = _mm256_mul_pd(beta,beta);
 443     beta3            = _mm256_mul_pd(beta,beta2);
 444
 445     ewtab            = fr->ic->tabq_coul_F;
 446     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 447     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 448
 449     /* Avoid stupid compiler warnings */
 450     jnrA = jnrB = jnrC = jnrD = 0;
 451     j_coord_offsetA = 0;
 452     j_coord_offsetB = 0;
 453     j_coord_offsetC = 0;
 454     j_coord_offsetD = 0;
 455
 456     outeriter        = 0;
 457     inneriter        = 0;
 458
 459     for(iidx=0;iidx<4*DIM;iidx++)
 460     {
 461         scratch[iidx] = 0.0;
 462     }
 463
 464     /* Start outer loop over neighborlists */
 465     for(iidx=0; iidx<nri; iidx++)
 466     {
 467         /* Load shift vector for this list */
 468         i_shift_offset   = DIM*shiftidx[iidx];
 469
 470         /* Load limits for loop over neighbors */
 471         j_index_start    = jindex[iidx];
 472         j_index_end      = jindex[iidx+1];
 473
 474         /* Get outer coordinate index */
 475         inr              = iinr[iidx];
 476         i_coord_offset   = DIM*inr;
 477
 478         /* Load i particle coords and add shift vector */
 479         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 480
 481         fix0             = _mm256_setzero_pd();
 482         fiy0             = _mm256_setzero_pd();
 483         fiz0             = _mm256_setzero_pd();
 484
 485         /* Load parameters for i particles */
 486         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 487
 488         /* Start inner kernel loop */
 489         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 490         {
 491
 492             /* Get j neighbor index, and coordinate index */
 493             jnrA             = jjnr[jidx];
 494             jnrB             = jjnr[jidx+1];
 495             jnrC             = jjnr[jidx+2];
 496             jnrD             = jjnr[jidx+3];
 497             j_coord_offsetA  = DIM*jnrA;
 498             j_coord_offsetB  = DIM*jnrB;
 499             j_coord_offsetC  = DIM*jnrC;
 500             j_coord_offsetD  = DIM*jnrD;
 501
 502             /* load j atom coordinates */
 503             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 504                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 505                                                  &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm256_sub_pd(ix0,jx0);
 509             dy00             = _mm256_sub_pd(iy0,jy0);
 510             dz00             = _mm256_sub_pd(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 516
 517             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 521                                                                  charge+jnrC+0,charge+jnrD+0);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r00              = _mm256_mul_pd(rsq00,rinv00);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq00             = _mm256_mul_pd(iq0,jq0);
 531
 532             /* EWALD ELECTROSTATICS */
 533
 534             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 535             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 536             ewitab           = _mm256_cvttpd_epi32(ewrt);
 537             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 538             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 539                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 540                                             &ewtabF,&ewtabFn);
 541             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 542             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 543
 544             fscal            = felec;
 545
 546             /* Calculate temporary vectorial force */
 547             tx               = _mm256_mul_pd(fscal,dx00);
 548             ty               = _mm256_mul_pd(fscal,dy00);
 549             tz               = _mm256_mul_pd(fscal,dz00);
 550
 551             /* Update vectorial force */
 552             fix0             = _mm256_add_pd(fix0,tx);
 553             fiy0             = _mm256_add_pd(fiy0,ty);
 554             fiz0             = _mm256_add_pd(fiz0,tz);
 555
 556             fjptrA             = f+j_coord_offsetA;
 557             fjptrB             = f+j_coord_offsetB;
 558             fjptrC             = f+j_coord_offsetC;
 559             fjptrD             = f+j_coord_offsetD;
 560             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 561
 562             /* Inner loop uses 36 flops */
 563         }
 564
 565         if(jidx<j_index_end)
 566         {
 567
 568             /* Get j neighbor index, and coordinate index */
 569             jnrlistA         = jjnr[jidx];
 570             jnrlistB         = jjnr[jidx+1];
 571             jnrlistC         = jjnr[jidx+2];
 572             jnrlistD         = jjnr[jidx+3];
 573             /* Sign of each element will be negative for non-real atoms.
 574              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 575              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 576              */
 577             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 578
 579             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 580             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 581             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 582
 583             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 584             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 585             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 586             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 587             j_coord_offsetA  = DIM*jnrA;
 588             j_coord_offsetB  = DIM*jnrB;
 589             j_coord_offsetC  = DIM*jnrC;
 590             j_coord_offsetD  = DIM*jnrD;
 591
 592             /* load j atom coordinates */
 593             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 594                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 595                                                  &jx0,&jy0,&jz0);
 596
 597             /* Calculate displacement vector */
 598             dx00             = _mm256_sub_pd(ix0,jx0);
 599             dy00             = _mm256_sub_pd(iy0,jy0);
 600             dz00             = _mm256_sub_pd(iz0,jz0);
 601
 602             /* Calculate squared distance and things based on it */
 603             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 604
 605             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 606
 607             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 608
 609             /* Load parameters for j particles */
 610             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 611                                                                  charge+jnrC+0,charge+jnrD+0);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r00              = _mm256_mul_pd(rsq00,rinv00);
 618             r00              = _mm256_andnot_pd(dummy_mask,r00);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm256_mul_pd(iq0,jq0);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 627             ewitab           = _mm256_cvttpd_epi32(ewrt);
 628             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 629             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 630                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 631                                             &ewtabF,&ewtabFn);
 632             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 633             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 634
 635             fscal            = felec;
 636
 637             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_pd(fscal,dx00);
 641             ty               = _mm256_mul_pd(fscal,dy00);
 642             tz               = _mm256_mul_pd(fscal,dz00);
 643
 644             /* Update vectorial force */
 645             fix0             = _mm256_add_pd(fix0,tx);
 646             fiy0             = _mm256_add_pd(fiy0,ty);
 647             fiz0             = _mm256_add_pd(fiz0,tz);
 648
 649             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 650             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 651             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 652             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 653             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 654
 655             /* Inner loop uses 37 flops */
 656         }
 657
 658         /* End of innermost loop */
 659
 660         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 661                                                  f+i_coord_offset,fshift+i_shift_offset);
 662
 663         /* Increment number of inner iterations */
 664         inneriter                  += j_index_end - j_index_start;
 665
 666         /* Outer loop uses 7 flops */
 667     }
 668
 669     /* Increment number of outer iterations */
 670     outeriter        += nri;
 671
 672     /* Update outer/inner flops */
 673
 674     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);
 675 }