src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwNone_GeomP1P1_sse4_1_single.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 sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse4_1_single
  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 SSE, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm_set1_ps(fr->ic->c_rf);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     for(iidx=0;iidx<4*DIM;iidx++)
 122     {
 123         scratch[iidx] = 0.0;
 124     }
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_ps();
 144         fiy0             = _mm_setzero_ps();
 145         fiz0             = _mm_setzero_ps();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_ps();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             jnrC             = jjnr[jidx+2];
 161             jnrD             = jjnr[jidx+3];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164             j_coord_offsetC  = DIM*jnrC;
 165             j_coord_offsetD  = DIM*jnrD;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               x+j_coord_offsetC,x+j_coord_offsetD,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_ps(ix0,jx0);
 174             dy00             = _mm_sub_ps(iy0,jy0);
 175             dz00             = _mm_sub_ps(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 179
 180             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 181
 182             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 186                                                               charge+jnrC+0,charge+jnrD+0);
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _mm_mul_ps(iq0,jq0);
 194
 195             /* REACTION-FIELD ELECTROSTATICS */
 196             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 197             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 198
 199             /* Update potential sum for this i atom from the interaction with this j atom. */
 200             velecsum         = _mm_add_ps(velecsum,velec);
 201
 202             fscal            = felec;
 203
 204             /* Calculate temporary vectorial force */
 205             tx               = _mm_mul_ps(fscal,dx00);
 206             ty               = _mm_mul_ps(fscal,dy00);
 207             tz               = _mm_mul_ps(fscal,dz00);
 208
 209             /* Update vectorial force */
 210             fix0             = _mm_add_ps(fix0,tx);
 211             fiy0             = _mm_add_ps(fiy0,ty);
 212             fiz0             = _mm_add_ps(fiz0,tz);
 213
 214             fjptrA             = f+j_coord_offsetA;
 215             fjptrB             = f+j_coord_offsetB;
 216             fjptrC             = f+j_coord_offsetC;
 217             fjptrD             = f+j_coord_offsetD;
 218             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 219
 220             /* Inner loop uses 32 flops */
 221         }
 222
 223         if(jidx<j_index_end)
 224         {
 225
 226             /* Get j neighbor index, and coordinate index */
 227             jnrlistA         = jjnr[jidx];
 228             jnrlistB         = jjnr[jidx+1];
 229             jnrlistC         = jjnr[jidx+2];
 230             jnrlistD         = jjnr[jidx+3];
 231             /* Sign of each element will be negative for non-real atoms.
 232              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 233              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 234              */
 235             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 236             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 237             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 238             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 239             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 240             j_coord_offsetA  = DIM*jnrA;
 241             j_coord_offsetB  = DIM*jnrB;
 242             j_coord_offsetC  = DIM*jnrC;
 243             j_coord_offsetD  = DIM*jnrD;
 244
 245             /* load j atom coordinates */
 246             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 247                                               x+j_coord_offsetC,x+j_coord_offsetD,
 248                                               &jx0,&jy0,&jz0);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm_sub_ps(ix0,jx0);
 252             dy00             = _mm_sub_ps(iy0,jy0);
 253             dz00             = _mm_sub_ps(iz0,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 257
 258             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 259
 260             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 261
 262             /* Load parameters for j particles */
 263             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 264                                                               charge+jnrC+0,charge+jnrD+0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm_mul_ps(iq0,jq0);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 275             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velec            = _mm_andnot_ps(dummy_mask,velec);
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280
 281             fscal            = felec;
 282
 283             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 296             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 297             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 298             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 299             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 300
 301             /* Inner loop uses 32 flops */
 302         }
 303
 304         /* End of innermost loop */
 305
 306         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 307                                               f+i_coord_offset,fshift+i_shift_offset);
 308
 309         ggid                        = gid[iidx];
 310         /* Update potential energies */
 311         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 312
 313         /* Increment number of inner iterations */
 314         inneriter                  += j_index_end - j_index_start;
 315
 316         /* Outer loop uses 8 flops */
 317     }
 318
 319     /* Increment number of outer iterations */
 320     outeriter        += nri;
 321
 322     /* Update outer/inner flops */
 323
 324     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 325 }
 326 /*
 327  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse4_1_single
 328  * Electrostatics interaction: ReactionField
 329  * VdW interaction:            None
 330  * Geometry:                   Particle-Particle
 331  * Calculate force/pot:        Force
 332  */
 333 void
 334 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse4_1_single
 335                     (t_nblist                    * gmx_restrict       nlist,
 336                      rvec                        * gmx_restrict          xx,
 337                      rvec                        * gmx_restrict          ff,
 338                      t_forcerec                  * gmx_restrict          fr,
 339                      t_mdatoms                   * gmx_restrict     mdatoms,
 340                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 341                      t_nrnb                      * gmx_restrict        nrnb)
 342 {
 343     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 344      * just 0 for non-waters.
 345      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 346      * jnr indices corresponding to data put in the four positions in the SIMD register.
 347      */
 348     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 349     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 350     int              jnrA,jnrB,jnrC,jnrD;
 351     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 352     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 353     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 354     real             rcutoff_scalar;
 355     real             *shiftvec,*fshift,*x,*f;
 356     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 357     real             scratch[4*DIM];
 358     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 359     int              vdwioffset0;
 360     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 361     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 362     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 363     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 364     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 365     real             *charge;
 366     __m128           dummy_mask,cutoff_mask;
 367     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 368     __m128           one     = _mm_set1_ps(1.0);
 369     __m128           two     = _mm_set1_ps(2.0);
 370     x                = xx[0];
 371     f                = ff[0];
 372
 373     nri              = nlist->nri;
 374     iinr             = nlist->iinr;
 375     jindex           = nlist->jindex;
 376     jjnr             = nlist->jjnr;
 377     shiftidx         = nlist->shift;
 378     gid              = nlist->gid;
 379     shiftvec         = fr->shift_vec[0];
 380     fshift           = fr->fshift[0];
 381     facel            = _mm_set1_ps(fr->epsfac);
 382     charge           = mdatoms->chargeA;
 383     krf              = _mm_set1_ps(fr->ic->k_rf);
 384     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 385     crf              = _mm_set1_ps(fr->ic->c_rf);
 386
 387     /* Avoid stupid compiler warnings */
 388     jnrA = jnrB = jnrC = jnrD = 0;
 389     j_coord_offsetA = 0;
 390     j_coord_offsetB = 0;
 391     j_coord_offsetC = 0;
 392     j_coord_offsetD = 0;
 393
 394     outeriter        = 0;
 395     inneriter        = 0;
 396
 397     for(iidx=0;iidx<4*DIM;iidx++)
 398     {
 399         scratch[iidx] = 0.0;
 400     }
 401
 402     /* Start outer loop over neighborlists */
 403     for(iidx=0; iidx<nri; iidx++)
 404     {
 405         /* Load shift vector for this list */
 406         i_shift_offset   = DIM*shiftidx[iidx];
 407
 408         /* Load limits for loop over neighbors */
 409         j_index_start    = jindex[iidx];
 410         j_index_end      = jindex[iidx+1];
 411
 412         /* Get outer coordinate index */
 413         inr              = iinr[iidx];
 414         i_coord_offset   = DIM*inr;
 415
 416         /* Load i particle coords and add shift vector */
 417         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 418
 419         fix0             = _mm_setzero_ps();
 420         fiy0             = _mm_setzero_ps();
 421         fiz0             = _mm_setzero_ps();
 422
 423         /* Load parameters for i particles */
 424         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 425
 426         /* Start inner kernel loop */
 427         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 428         {
 429
 430             /* Get j neighbor index, and coordinate index */
 431             jnrA             = jjnr[jidx];
 432             jnrB             = jjnr[jidx+1];
 433             jnrC             = jjnr[jidx+2];
 434             jnrD             = jjnr[jidx+3];
 435             j_coord_offsetA  = DIM*jnrA;
 436             j_coord_offsetB  = DIM*jnrB;
 437             j_coord_offsetC  = DIM*jnrC;
 438             j_coord_offsetD  = DIM*jnrD;
 439
 440             /* load j atom coordinates */
 441             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 442                                               x+j_coord_offsetC,x+j_coord_offsetD,
 443                                               &jx0,&jy0,&jz0);
 444
 445             /* Calculate displacement vector */
 446             dx00             = _mm_sub_ps(ix0,jx0);
 447             dy00             = _mm_sub_ps(iy0,jy0);
 448             dz00             = _mm_sub_ps(iz0,jz0);
 449
 450             /* Calculate squared distance and things based on it */
 451             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 452
 453             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 454
 455             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 456
 457             /* Load parameters for j particles */
 458             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 459                                                               charge+jnrC+0,charge+jnrD+0);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq00             = _mm_mul_ps(iq0,jq0);
 467
 468             /* REACTION-FIELD ELECTROSTATICS */
 469             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 470
 471             fscal            = felec;
 472
 473             /* Calculate temporary vectorial force */
 474             tx               = _mm_mul_ps(fscal,dx00);
 475             ty               = _mm_mul_ps(fscal,dy00);
 476             tz               = _mm_mul_ps(fscal,dz00);
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_add_ps(fix0,tx);
 480             fiy0             = _mm_add_ps(fiy0,ty);
 481             fiz0             = _mm_add_ps(fiz0,tz);
 482
 483             fjptrA             = f+j_coord_offsetA;
 484             fjptrB             = f+j_coord_offsetB;
 485             fjptrC             = f+j_coord_offsetC;
 486             fjptrD             = f+j_coord_offsetD;
 487             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 488
 489             /* Inner loop uses 27 flops */
 490         }
 491
 492         if(jidx<j_index_end)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrlistA         = jjnr[jidx];
 497             jnrlistB         = jjnr[jidx+1];
 498             jnrlistC         = jjnr[jidx+2];
 499             jnrlistD         = jjnr[jidx+3];
 500             /* Sign of each element will be negative for non-real atoms.
 501              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 502              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 503              */
 504             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 505             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 506             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 507             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 508             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511             j_coord_offsetC  = DIM*jnrC;
 512             j_coord_offsetD  = DIM*jnrD;
 513
 514             /* load j atom coordinates */
 515             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 516                                               x+j_coord_offsetC,x+j_coord_offsetD,
 517                                               &jx0,&jy0,&jz0);
 518
 519             /* Calculate displacement vector */
 520             dx00             = _mm_sub_ps(ix0,jx0);
 521             dy00             = _mm_sub_ps(iy0,jy0);
 522             dz00             = _mm_sub_ps(iz0,jz0);
 523
 524             /* Calculate squared distance and things based on it */
 525             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 526
 527             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 528
 529             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 530
 531             /* Load parameters for j particles */
 532             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 533                                                               charge+jnrC+0,charge+jnrD+0);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq00             = _mm_mul_ps(iq0,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm_mul_ps(fscal,dx00);
 551             ty               = _mm_mul_ps(fscal,dy00);
 552             tz               = _mm_mul_ps(fscal,dz00);
 553
 554             /* Update vectorial force */
 555             fix0             = _mm_add_ps(fix0,tx);
 556             fiy0             = _mm_add_ps(fiy0,ty);
 557             fiz0             = _mm_add_ps(fiz0,tz);
 558
 559             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 560             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 561             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 562             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 563             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 564
 565             /* Inner loop uses 27 flops */
 566         }
 567
 568         /* End of innermost loop */
 569
 570         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 571                                               f+i_coord_offset,fshift+i_shift_offset);
 572
 573         /* Increment number of inner iterations */
 574         inneriter                  += j_index_end - j_index_start;
 575
 576         /* Outer loop uses 7 flops */
 577     }
 578
 579     /* Increment number of outer iterations */
 580     outeriter        += nri;
 581
 582     /* Update outer/inner flops */
 583
 584     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 585 }