src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwNone_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_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     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m256d          dummy_mask,cutoff_mask;
  96     __m128           tmpmask0,tmpmask1;
  97     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  98     __m256d          one     = _mm256_set1_pd(1.0);
  99     __m256d          two     = _mm256_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_pd(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113
 114     vftab            = kernel_data->table_elec->data;
 115     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm256_setzero_pd();
 150         fiy0             = _mm256_setzero_pd();
 151         fiz0             = _mm256_setzero_pd();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm256_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 176                                                  &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm256_sub_pd(ix0,jx0);
 180             dy00             = _mm256_sub_pd(iy0,jy0);
 181             dz00             = _mm256_sub_pd(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 190                                                                  charge+jnrC+0,charge+jnrD+0);
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm256_mul_pd(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm256_mul_pd(iq0,jq0);
 200
 201             /* Calculate table index by multiplying r with table scale and truncate to integer */
 202             rt               = _mm256_mul_pd(r00,vftabscale);
 203             vfitab           = _mm256_cvttpd_epi32(rt);
 204             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 205             vfitab           = _mm_slli_epi32(vfitab,2);
 206
 207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 208             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 209             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 210             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 211             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 212             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 213             Heps             = _mm256_mul_pd(vfeps,H);
 214             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 215             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 216             velec            = _mm256_mul_pd(qq00,VV);
 217             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 218             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             velecsum         = _mm256_add_pd(velecsum,velec);
 222
 223             fscal            = felec;
 224
 225             /* Calculate temporary vectorial force */
 226             tx               = _mm256_mul_pd(fscal,dx00);
 227             ty               = _mm256_mul_pd(fscal,dy00);
 228             tz               = _mm256_mul_pd(fscal,dz00);
 229
 230             /* Update vectorial force */
 231             fix0             = _mm256_add_pd(fix0,tx);
 232             fiy0             = _mm256_add_pd(fiy0,ty);
 233             fiz0             = _mm256_add_pd(fiz0,tz);
 234
 235             fjptrA             = f+j_coord_offsetA;
 236             fjptrB             = f+j_coord_offsetB;
 237             fjptrC             = f+j_coord_offsetC;
 238             fjptrD             = f+j_coord_offsetD;
 239             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 240
 241             /* Inner loop uses 43 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             /* Get j neighbor index, and coordinate index */
 248             jnrlistA         = jjnr[jidx];
 249             jnrlistB         = jjnr[jidx+1];
 250             jnrlistC         = jjnr[jidx+2];
 251             jnrlistD         = jjnr[jidx+3];
 252             /* Sign of each element will be negative for non-real atoms.
 253              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 254              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 255              */
 256             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 257
 258             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 259             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 260             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 261
 262             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 263             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 264             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 265             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 266             j_coord_offsetA  = DIM*jnrA;
 267             j_coord_offsetB  = DIM*jnrB;
 268             j_coord_offsetC  = DIM*jnrC;
 269             j_coord_offsetD  = DIM*jnrD;
 270
 271             /* load j atom coordinates */
 272             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 273                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 274                                                  &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _mm256_sub_pd(ix0,jx0);
 278             dy00             = _mm256_sub_pd(iy0,jy0);
 279             dz00             = _mm256_sub_pd(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 283
 284             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 288                                                                  charge+jnrC+0,charge+jnrD+0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r00              = _mm256_mul_pd(rsq00,rinv00);
 295             r00              = _mm256_andnot_pd(dummy_mask,r00);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm256_mul_pd(iq0,jq0);
 299
 300             /* Calculate table index by multiplying r with table scale and truncate to integer */
 301             rt               = _mm256_mul_pd(r00,vftabscale);
 302             vfitab           = _mm256_cvttpd_epi32(rt);
 303             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 304             vfitab           = _mm_slli_epi32(vfitab,2);
 305
 306             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 307             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 308             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 309             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 310             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 311             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 312             Heps             = _mm256_mul_pd(vfeps,H);
 313             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 314             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 315             velec            = _mm256_mul_pd(qq00,VV);
 316             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 317             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm256_andnot_pd(dummy_mask,velec);
 321             velecsum         = _mm256_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_pd(fscal,dx00);
 329             ty               = _mm256_mul_pd(fscal,dy00);
 330             tz               = _mm256_mul_pd(fscal,dz00);
 331
 332             /* Update vectorial force */
 333             fix0             = _mm256_add_pd(fix0,tx);
 334             fiy0             = _mm256_add_pd(fiy0,ty);
 335             fiz0             = _mm256_add_pd(fiz0,tz);
 336
 337             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 338             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 339             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 340             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 341             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 342
 343             /* Inner loop uses 44 flops */
 344         }
 345
 346         /* End of innermost loop */
 347
 348         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 349                                                  f+i_coord_offset,fshift+i_shift_offset);
 350
 351         ggid                        = gid[iidx];
 352         /* Update potential energies */
 353         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 354
 355         /* Increment number of inner iterations */
 356         inneriter                  += j_index_end - j_index_start;
 357
 358         /* Outer loop uses 8 flops */
 359     }
 360
 361     /* Increment number of outer iterations */
 362     outeriter        += nri;
 363
 364     /* Update outer/inner flops */
 365
 366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 367 }
 368 /*
 369  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_256_double
 370  * Electrostatics interaction: CubicSplineTable
 371  * VdW interaction:            None
 372  * Geometry:                   Particle-Particle
 373  * Calculate force/pot:        Force
 374  */
 375 void
 376 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_256_double
 377                     (t_nblist                    * gmx_restrict       nlist,
 378                      rvec                        * gmx_restrict          xx,
 379                      rvec                        * gmx_restrict          ff,
 380                      t_forcerec                  * gmx_restrict          fr,
 381                      t_mdatoms                   * gmx_restrict     mdatoms,
 382                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 383                      t_nrnb                      * gmx_restrict        nrnb)
 384 {
 385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 386      * just 0 for non-waters.
 387      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 388      * jnr indices corresponding to data put in the four positions in the SIMD register.
 389      */
 390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 392     int              jnrA,jnrB,jnrC,jnrD;
 393     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 394     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 395     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 396     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 397     real             rcutoff_scalar;
 398     real             *shiftvec,*fshift,*x,*f;
 399     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 400     real             scratch[4*DIM];
 401     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 402     real *           vdwioffsetptr0;
 403     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 404     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 405     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 406     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 407     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 408     real             *charge;
 409     __m128i          vfitab;
 410     __m128i          ifour       = _mm_set1_epi32(4);
 411     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 412     real             *vftab;
 413     __m256d          dummy_mask,cutoff_mask;
 414     __m128           tmpmask0,tmpmask1;
 415     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 416     __m256d          one     = _mm256_set1_pd(1.0);
 417     __m256d          two     = _mm256_set1_pd(2.0);
 418     x                = xx[0];
 419     f                = ff[0];
 420
 421     nri              = nlist->nri;
 422     iinr             = nlist->iinr;
 423     jindex           = nlist->jindex;
 424     jjnr             = nlist->jjnr;
 425     shiftidx         = nlist->shift;
 426     gid              = nlist->gid;
 427     shiftvec         = fr->shift_vec[0];
 428     fshift           = fr->fshift[0];
 429     facel            = _mm256_set1_pd(fr->epsfac);
 430     charge           = mdatoms->chargeA;
 431
 432     vftab            = kernel_data->table_elec->data;
 433     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 434
 435     /* Avoid stupid compiler warnings */
 436     jnrA = jnrB = jnrC = jnrD = 0;
 437     j_coord_offsetA = 0;
 438     j_coord_offsetB = 0;
 439     j_coord_offsetC = 0;
 440     j_coord_offsetD = 0;
 441
 442     outeriter        = 0;
 443     inneriter        = 0;
 444
 445     for(iidx=0;iidx<4*DIM;iidx++)
 446     {
 447         scratch[iidx] = 0.0;
 448     }
 449
 450     /* Start outer loop over neighborlists */
 451     for(iidx=0; iidx<nri; iidx++)
 452     {
 453         /* Load shift vector for this list */
 454         i_shift_offset   = DIM*shiftidx[iidx];
 455
 456         /* Load limits for loop over neighbors */
 457         j_index_start    = jindex[iidx];
 458         j_index_end      = jindex[iidx+1];
 459
 460         /* Get outer coordinate index */
 461         inr              = iinr[iidx];
 462         i_coord_offset   = DIM*inr;
 463
 464         /* Load i particle coords and add shift vector */
 465         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 466
 467         fix0             = _mm256_setzero_pd();
 468         fiy0             = _mm256_setzero_pd();
 469         fiz0             = _mm256_setzero_pd();
 470
 471         /* Load parameters for i particles */
 472         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 473
 474         /* Start inner kernel loop */
 475         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 476         {
 477
 478             /* Get j neighbor index, and coordinate index */
 479             jnrA             = jjnr[jidx];
 480             jnrB             = jjnr[jidx+1];
 481             jnrC             = jjnr[jidx+2];
 482             jnrD             = jjnr[jidx+3];
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487
 488             /* load j atom coordinates */
 489             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 491                                                  &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm256_sub_pd(ix0,jx0);
 495             dy00             = _mm256_sub_pd(iy0,jy0);
 496             dz00             = _mm256_sub_pd(iz0,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 500
 501             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 502
 503             /* Load parameters for j particles */
 504             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 505                                                                  charge+jnrC+0,charge+jnrD+0);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r00              = _mm256_mul_pd(rsq00,rinv00);
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq00             = _mm256_mul_pd(iq0,jq0);
 515
 516             /* Calculate table index by multiplying r with table scale and truncate to integer */
 517             rt               = _mm256_mul_pd(r00,vftabscale);
 518             vfitab           = _mm256_cvttpd_epi32(rt);
 519             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 520             vfitab           = _mm_slli_epi32(vfitab,2);
 521
 522             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 523             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 524             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 525             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 526             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 527             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 528             Heps             = _mm256_mul_pd(vfeps,H);
 529             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 530             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 531             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 532
 533             fscal            = felec;
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_pd(fscal,dx00);
 537             ty               = _mm256_mul_pd(fscal,dy00);
 538             tz               = _mm256_mul_pd(fscal,dz00);
 539
 540             /* Update vectorial force */
 541             fix0             = _mm256_add_pd(fix0,tx);
 542             fiy0             = _mm256_add_pd(fiy0,ty);
 543             fiz0             = _mm256_add_pd(fiz0,tz);
 544
 545             fjptrA             = f+j_coord_offsetA;
 546             fjptrB             = f+j_coord_offsetB;
 547             fjptrC             = f+j_coord_offsetC;
 548             fjptrD             = f+j_coord_offsetD;
 549             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 550
 551             /* Inner loop uses 39 flops */
 552         }
 553
 554         if(jidx<j_index_end)
 555         {
 556
 557             /* Get j neighbor index, and coordinate index */
 558             jnrlistA         = jjnr[jidx];
 559             jnrlistB         = jjnr[jidx+1];
 560             jnrlistC         = jjnr[jidx+2];
 561             jnrlistD         = jjnr[jidx+3];
 562             /* Sign of each element will be negative for non-real atoms.
 563              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 564              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 565              */
 566             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 567
 568             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 569             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 570             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 571
 572             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 573             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 574             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 575             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 576             j_coord_offsetA  = DIM*jnrA;
 577             j_coord_offsetB  = DIM*jnrB;
 578             j_coord_offsetC  = DIM*jnrC;
 579             j_coord_offsetD  = DIM*jnrD;
 580
 581             /* load j atom coordinates */
 582             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 583                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 584                                                  &jx0,&jy0,&jz0);
 585
 586             /* Calculate displacement vector */
 587             dx00             = _mm256_sub_pd(ix0,jx0);
 588             dy00             = _mm256_sub_pd(iy0,jy0);
 589             dz00             = _mm256_sub_pd(iz0,jz0);
 590
 591             /* Calculate squared distance and things based on it */
 592             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 593
 594             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 598                                                                  charge+jnrC+0,charge+jnrD+0);
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             r00              = _mm256_mul_pd(rsq00,rinv00);
 605             r00              = _mm256_andnot_pd(dummy_mask,r00);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq00             = _mm256_mul_pd(iq0,jq0);
 609
 610             /* Calculate table index by multiplying r with table scale and truncate to integer */
 611             rt               = _mm256_mul_pd(r00,vftabscale);
 612             vfitab           = _mm256_cvttpd_epi32(rt);
 613             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 614             vfitab           = _mm_slli_epi32(vfitab,2);
 615
 616             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 617             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 618             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 619             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 620             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 621             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 622             Heps             = _mm256_mul_pd(vfeps,H);
 623             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 624             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 625             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_pd(fscal,dx00);
 633             ty               = _mm256_mul_pd(fscal,dy00);
 634             tz               = _mm256_mul_pd(fscal,dz00);
 635
 636             /* Update vectorial force */
 637             fix0             = _mm256_add_pd(fix0,tx);
 638             fiy0             = _mm256_add_pd(fiy0,ty);
 639             fiz0             = _mm256_add_pd(fiz0,tz);
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 646
 647             /* Inner loop uses 40 flops */
 648         }
 649
 650         /* End of innermost loop */
 651
 652         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 653                                                  f+i_coord_offset,fshift+i_shift_offset);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 7 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 667 }