src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     __m256d          dummy_mask,cutoff_mask;
  92     __m128           tmpmask0,tmpmask1;
  93     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  94     __m256d          one     = _mm256_set1_pd(1.0);
  95     __m256d          two     = _mm256_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm256_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = jnrC = jnrD = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114     j_coord_offsetC = 0;
 115     j_coord_offsetD = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     for(iidx=0;iidx<4*DIM;iidx++)
 121     {
 122         scratch[iidx] = 0.0;
 123     }
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm256_setzero_pd();
 143         fiy0             = _mm256_setzero_pd();
 144         fiz0             = _mm256_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm256_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             jnrC             = jjnr[jidx+2];
 160             jnrD             = jjnr[jidx+3];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 169                                                  &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm256_sub_pd(ix0,jx0);
 173             dy00             = _mm256_sub_pd(iy0,jy0);
 174             dz00             = _mm256_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 180
 181             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 185                                                                  charge+jnrC+0,charge+jnrD+0);
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm256_mul_pd(iq0,jq0);
 193
 194             /* COULOMB ELECTROSTATICS */
 195             velec            = _mm256_mul_pd(qq00,rinv00);
 196             felec            = _mm256_mul_pd(velec,rinvsq00);
 197
 198             /* Update potential sum for this i atom from the interaction with this j atom. */
 199             velecsum         = _mm256_add_pd(velecsum,velec);
 200
 201             fscal            = felec;
 202
 203             /* Calculate temporary vectorial force */
 204             tx               = _mm256_mul_pd(fscal,dx00);
 205             ty               = _mm256_mul_pd(fscal,dy00);
 206             tz               = _mm256_mul_pd(fscal,dz00);
 207
 208             /* Update vectorial force */
 209             fix0             = _mm256_add_pd(fix0,tx);
 210             fiy0             = _mm256_add_pd(fiy0,ty);
 211             fiz0             = _mm256_add_pd(fiz0,tz);
 212
 213             fjptrA             = f+j_coord_offsetA;
 214             fjptrB             = f+j_coord_offsetB;
 215             fjptrC             = f+j_coord_offsetC;
 216             fjptrD             = f+j_coord_offsetD;
 217             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 218
 219             /* Inner loop uses 27 flops */
 220         }
 221
 222         if(jidx<j_index_end)
 223         {
 224
 225             /* Get j neighbor index, and coordinate index */
 226             jnrlistA         = jjnr[jidx];
 227             jnrlistB         = jjnr[jidx+1];
 228             jnrlistC         = jjnr[jidx+2];
 229             jnrlistD         = jjnr[jidx+3];
 230             /* Sign of each element will be negative for non-real atoms.
 231              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 232              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 233              */
 234             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 235
 236             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 237             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 238             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 239
 240             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 241             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 242             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 243             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 244             j_coord_offsetA  = DIM*jnrA;
 245             j_coord_offsetB  = DIM*jnrB;
 246             j_coord_offsetC  = DIM*jnrC;
 247             j_coord_offsetD  = DIM*jnrD;
 248
 249             /* load j atom coordinates */
 250             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 251                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 252                                                  &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _mm256_sub_pd(ix0,jx0);
 256             dy00             = _mm256_sub_pd(iy0,jy0);
 257             dz00             = _mm256_sub_pd(iz0,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 261
 262             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 263
 264             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 268                                                                  charge+jnrC+0,charge+jnrD+0);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq00             = _mm256_mul_pd(iq0,jq0);
 276
 277             /* COULOMB ELECTROSTATICS */
 278             velec            = _mm256_mul_pd(qq00,rinv00);
 279             felec            = _mm256_mul_pd(velec,rinvsq00);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm256_andnot_pd(dummy_mask,velec);
 283             velecsum         = _mm256_add_pd(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_pd(fscal,dx00);
 291             ty               = _mm256_mul_pd(fscal,dy00);
 292             tz               = _mm256_mul_pd(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_pd(fix0,tx);
 296             fiy0             = _mm256_add_pd(fiy0,ty);
 297             fiz0             = _mm256_add_pd(fiz0,tz);
 298
 299             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 300             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 301             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 302             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 303             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 304
 305             /* Inner loop uses 27 flops */
 306         }
 307
 308         /* End of innermost loop */
 309
 310         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 311                                                  f+i_coord_offset,fshift+i_shift_offset);
 312
 313         ggid                        = gid[iidx];
 314         /* Update potential energies */
 315         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 316
 317         /* Increment number of inner iterations */
 318         inneriter                  += j_index_end - j_index_start;
 319
 320         /* Outer loop uses 8 flops */
 321     }
 322
 323     /* Increment number of outer iterations */
 324     outeriter        += nri;
 325
 326     /* Update outer/inner flops */
 327
 328     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 329 }
 330 /*
 331  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 332  * Electrostatics interaction: Coulomb
 333  * VdW interaction:            None
 334  * Geometry:                   Particle-Particle
 335  * Calculate force/pot:        Force
 336  */
 337 void
 338 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 339                     (t_nblist                    * gmx_restrict       nlist,
 340                      rvec                        * gmx_restrict          xx,
 341                      rvec                        * gmx_restrict          ff,
 342                      t_forcerec                  * gmx_restrict          fr,
 343                      t_mdatoms                   * gmx_restrict     mdatoms,
 344                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 345                      t_nrnb                      * gmx_restrict        nrnb)
 346 {
 347     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 348      * just 0 for non-waters.
 349      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 350      * jnr indices corresponding to data put in the four positions in the SIMD register.
 351      */
 352     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 353     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 354     int              jnrA,jnrB,jnrC,jnrD;
 355     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 356     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 357     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 358     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 359     real             rcutoff_scalar;
 360     real             *shiftvec,*fshift,*x,*f;
 361     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 362     real             scratch[4*DIM];
 363     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 364     real *           vdwioffsetptr0;
 365     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 366     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 367     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 368     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 369     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 370     real             *charge;
 371     __m256d          dummy_mask,cutoff_mask;
 372     __m128           tmpmask0,tmpmask1;
 373     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 374     __m256d          one     = _mm256_set1_pd(1.0);
 375     __m256d          two     = _mm256_set1_pd(2.0);
 376     x                = xx[0];
 377     f                = ff[0];
 378
 379     nri              = nlist->nri;
 380     iinr             = nlist->iinr;
 381     jindex           = nlist->jindex;
 382     jjnr             = nlist->jjnr;
 383     shiftidx         = nlist->shift;
 384     gid              = nlist->gid;
 385     shiftvec         = fr->shift_vec[0];
 386     fshift           = fr->fshift[0];
 387     facel            = _mm256_set1_pd(fr->epsfac);
 388     charge           = mdatoms->chargeA;
 389
 390     /* Avoid stupid compiler warnings */
 391     jnrA = jnrB = jnrC = jnrD = 0;
 392     j_coord_offsetA = 0;
 393     j_coord_offsetB = 0;
 394     j_coord_offsetC = 0;
 395     j_coord_offsetD = 0;
 396
 397     outeriter        = 0;
 398     inneriter        = 0;
 399
 400     for(iidx=0;iidx<4*DIM;iidx++)
 401     {
 402         scratch[iidx] = 0.0;
 403     }
 404
 405     /* Start outer loop over neighborlists */
 406     for(iidx=0; iidx<nri; iidx++)
 407     {
 408         /* Load shift vector for this list */
 409         i_shift_offset   = DIM*shiftidx[iidx];
 410
 411         /* Load limits for loop over neighbors */
 412         j_index_start    = jindex[iidx];
 413         j_index_end      = jindex[iidx+1];
 414
 415         /* Get outer coordinate index */
 416         inr              = iinr[iidx];
 417         i_coord_offset   = DIM*inr;
 418
 419         /* Load i particle coords and add shift vector */
 420         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 421
 422         fix0             = _mm256_setzero_pd();
 423         fiy0             = _mm256_setzero_pd();
 424         fiz0             = _mm256_setzero_pd();
 425
 426         /* Load parameters for i particles */
 427         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 428
 429         /* Start inner kernel loop */
 430         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 431         {
 432
 433             /* Get j neighbor index, and coordinate index */
 434             jnrA             = jjnr[jidx];
 435             jnrB             = jjnr[jidx+1];
 436             jnrC             = jjnr[jidx+2];
 437             jnrD             = jjnr[jidx+3];
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442
 443             /* load j atom coordinates */
 444             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 445                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 446                                                  &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm256_sub_pd(ix0,jx0);
 450             dy00             = _mm256_sub_pd(iy0,jy0);
 451             dz00             = _mm256_sub_pd(iz0,jz0);
 452
 453             /* Calculate squared distance and things based on it */
 454             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 455
 456             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 457
 458             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 459
 460             /* Load parameters for j particles */
 461             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 462                                                                  charge+jnrC+0,charge+jnrD+0);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm256_mul_pd(iq0,jq0);
 470
 471             /* COULOMB ELECTROSTATICS */
 472             velec            = _mm256_mul_pd(qq00,rinv00);
 473             felec            = _mm256_mul_pd(velec,rinvsq00);
 474
 475             fscal            = felec;
 476
 477             /* Calculate temporary vectorial force */
 478             tx               = _mm256_mul_pd(fscal,dx00);
 479             ty               = _mm256_mul_pd(fscal,dy00);
 480             tz               = _mm256_mul_pd(fscal,dz00);
 481
 482             /* Update vectorial force */
 483             fix0             = _mm256_add_pd(fix0,tx);
 484             fiy0             = _mm256_add_pd(fiy0,ty);
 485             fiz0             = _mm256_add_pd(fiz0,tz);
 486
 487             fjptrA             = f+j_coord_offsetA;
 488             fjptrB             = f+j_coord_offsetB;
 489             fjptrC             = f+j_coord_offsetC;
 490             fjptrD             = f+j_coord_offsetD;
 491             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 492
 493             /* Inner loop uses 26 flops */
 494         }
 495
 496         if(jidx<j_index_end)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrlistA         = jjnr[jidx];
 501             jnrlistB         = jjnr[jidx+1];
 502             jnrlistC         = jjnr[jidx+2];
 503             jnrlistD         = jjnr[jidx+3];
 504             /* Sign of each element will be negative for non-real atoms.
 505              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 506              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 507              */
 508             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 509
 510             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 511             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 512             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 513
 514             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 515             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 516             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 517             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 518             j_coord_offsetA  = DIM*jnrA;
 519             j_coord_offsetB  = DIM*jnrB;
 520             j_coord_offsetC  = DIM*jnrC;
 521             j_coord_offsetD  = DIM*jnrD;
 522
 523             /* load j atom coordinates */
 524             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 525                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 526                                                  &jx0,&jy0,&jz0);
 527
 528             /* Calculate displacement vector */
 529             dx00             = _mm256_sub_pd(ix0,jx0);
 530             dy00             = _mm256_sub_pd(iy0,jy0);
 531             dz00             = _mm256_sub_pd(iz0,jz0);
 532
 533             /* Calculate squared distance and things based on it */
 534             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 535
 536             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 537
 538             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 539
 540             /* Load parameters for j particles */
 541             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 542                                                                  charge+jnrC+0,charge+jnrD+0);
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm256_mul_pd(iq0,jq0);
 550
 551             /* COULOMB ELECTROSTATICS */
 552             velec            = _mm256_mul_pd(qq00,rinv00);
 553             felec            = _mm256_mul_pd(velec,rinvsq00);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_pd(fscal,dx00);
 561             ty               = _mm256_mul_pd(fscal,dy00);
 562             tz               = _mm256_mul_pd(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm256_add_pd(fix0,tx);
 566             fiy0             = _mm256_add_pd(fiy0,ty);
 567             fiz0             = _mm256_add_pd(fiz0,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 574
 575             /* Inner loop uses 26 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 581                                                  f+i_coord_offset,fshift+i_shift_offset);
 582
 583         /* Increment number of inner iterations */
 584         inneriter                  += j_index_end - j_index_start;
 585
 586         /* Outer loop uses 7 flops */
 587     }
 588
 589     /* Increment number of outer iterations */
 590     outeriter        += nri;
 591
 592     /* Update outer/inner flops */
 593
 594     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 595 }