src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwNone_GeomP1P1_sse2_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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_single
  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 SSE, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(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            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_ps();
 146         fiy0             = _mm_setzero_ps();
 147         fiz0             = _mm_setzero_ps();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 183
 184             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 188                                                               charge+jnrC+0,charge+jnrD+0);
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_ps(iq0,jq0);
 196
 197             /* REACTION-FIELD ELECTROSTATICS */
 198             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 199             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 200
 201             /* Update potential sum for this i atom from the interaction with this j atom. */
 202             velecsum         = _mm_add_ps(velecsum,velec);
 203
 204             fscal            = felec;
 205
 206             /* Calculate temporary vectorial force */
 207             tx               = _mm_mul_ps(fscal,dx00);
 208             ty               = _mm_mul_ps(fscal,dy00);
 209             tz               = _mm_mul_ps(fscal,dz00);
 210
 211             /* Update vectorial force */
 212             fix0             = _mm_add_ps(fix0,tx);
 213             fiy0             = _mm_add_ps(fiy0,ty);
 214             fiz0             = _mm_add_ps(fiz0,tz);
 215
 216             fjptrA             = f+j_coord_offsetA;
 217             fjptrB             = f+j_coord_offsetB;
 218             fjptrC             = f+j_coord_offsetC;
 219             fjptrD             = f+j_coord_offsetD;
 220             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 221
 222             /* Inner loop uses 32 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             /* Get j neighbor index, and coordinate index */
 229             jnrlistA         = jjnr[jidx];
 230             jnrlistB         = jjnr[jidx+1];
 231             jnrlistC         = jjnr[jidx+2];
 232             jnrlistD         = jjnr[jidx+3];
 233             /* Sign of each element will be negative for non-real atoms.
 234              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 235              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 236              */
 237             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 238             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 239             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 240             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 241             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 242             j_coord_offsetA  = DIM*jnrA;
 243             j_coord_offsetB  = DIM*jnrB;
 244             j_coord_offsetC  = DIM*jnrC;
 245             j_coord_offsetD  = DIM*jnrD;
 246
 247             /* load j atom coordinates */
 248             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 249                                               x+j_coord_offsetC,x+j_coord_offsetD,
 250                                               &jx0,&jy0,&jz0);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _mm_sub_ps(ix0,jx0);
 254             dy00             = _mm_sub_ps(iy0,jy0);
 255             dz00             = _mm_sub_ps(iz0,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 259
 260             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 261
 262             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 263
 264             /* Load parameters for j particles */
 265             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 266                                                               charge+jnrC+0,charge+jnrD+0);
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm_mul_ps(iq0,jq0);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 277             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velec            = _mm_andnot_ps(dummy_mask,velec);
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282
 283             fscal            = felec;
 284
 285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_ps(fscal,dx00);
 289             ty               = _mm_mul_ps(fscal,dy00);
 290             tz               = _mm_mul_ps(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_ps(fix0,tx);
 294             fiy0             = _mm_add_ps(fiy0,ty);
 295             fiz0             = _mm_add_ps(fiz0,tz);
 296
 297             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 298             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 299             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 300             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 301             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 302
 303             /* Inner loop uses 32 flops */
 304         }
 305
 306         /* End of innermost loop */
 307
 308         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 309                                               f+i_coord_offset,fshift+i_shift_offset);
 310
 311         ggid                        = gid[iidx];
 312         /* Update potential energies */
 313         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 314
 315         /* Increment number of inner iterations */
 316         inneriter                  += j_index_end - j_index_start;
 317
 318         /* Outer loop uses 8 flops */
 319     }
 320
 321     /* Increment number of outer iterations */
 322     outeriter        += nri;
 323
 324     /* Update outer/inner flops */
 325
 326     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 327 }
 328 /*
 329  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_single
 330  * Electrostatics interaction: ReactionField
 331  * VdW interaction:            None
 332  * Geometry:                   Particle-Particle
 333  * Calculate force/pot:        Force
 334  */
 335 void
 336 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_single
 337                     (t_nblist                    * gmx_restrict       nlist,
 338                      rvec                        * gmx_restrict          xx,
 339                      rvec                        * gmx_restrict          ff,
 340                      t_forcerec                  * gmx_restrict          fr,
 341                      t_mdatoms                   * gmx_restrict     mdatoms,
 342                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 343                      t_nrnb                      * gmx_restrict        nrnb)
 344 {
 345     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 346      * just 0 for non-waters.
 347      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 348      * jnr indices corresponding to data put in the four positions in the SIMD register.
 349      */
 350     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 351     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 352     int              jnrA,jnrB,jnrC,jnrD;
 353     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 354     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 355     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 356     real             rcutoff_scalar;
 357     real             *shiftvec,*fshift,*x,*f;
 358     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 359     real             scratch[4*DIM];
 360     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 361     int              vdwioffset0;
 362     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 363     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 364     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 365     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 366     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 367     real             *charge;
 368     __m128           dummy_mask,cutoff_mask;
 369     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 370     __m128           one     = _mm_set1_ps(1.0);
 371     __m128           two     = _mm_set1_ps(2.0);
 372     x                = xx[0];
 373     f                = ff[0];
 374
 375     nri              = nlist->nri;
 376     iinr             = nlist->iinr;
 377     jindex           = nlist->jindex;
 378     jjnr             = nlist->jjnr;
 379     shiftidx         = nlist->shift;
 380     gid              = nlist->gid;
 381     shiftvec         = fr->shift_vec[0];
 382     fshift           = fr->fshift[0];
 383     facel            = _mm_set1_ps(fr->epsfac);
 384     charge           = mdatoms->chargeA;
 385     krf              = _mm_set1_ps(fr->ic->k_rf);
 386     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 387     crf              = _mm_set1_ps(fr->ic->c_rf);
 388
 389     /* Avoid stupid compiler warnings */
 390     jnrA = jnrB = jnrC = jnrD = 0;
 391     j_coord_offsetA = 0;
 392     j_coord_offsetB = 0;
 393     j_coord_offsetC = 0;
 394     j_coord_offsetD = 0;
 395
 396     outeriter        = 0;
 397     inneriter        = 0;
 398
 399     for(iidx=0;iidx<4*DIM;iidx++)
 400     {
 401         scratch[iidx] = 0.0;
 402     }
 403
 404     /* Start outer loop over neighborlists */
 405     for(iidx=0; iidx<nri; iidx++)
 406     {
 407         /* Load shift vector for this list */
 408         i_shift_offset   = DIM*shiftidx[iidx];
 409
 410         /* Load limits for loop over neighbors */
 411         j_index_start    = jindex[iidx];
 412         j_index_end      = jindex[iidx+1];
 413
 414         /* Get outer coordinate index */
 415         inr              = iinr[iidx];
 416         i_coord_offset   = DIM*inr;
 417
 418         /* Load i particle coords and add shift vector */
 419         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 420
 421         fix0             = _mm_setzero_ps();
 422         fiy0             = _mm_setzero_ps();
 423         fiz0             = _mm_setzero_ps();
 424
 425         /* Load parameters for i particles */
 426         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 427
 428         /* Start inner kernel loop */
 429         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 430         {
 431
 432             /* Get j neighbor index, and coordinate index */
 433             jnrA             = jjnr[jidx];
 434             jnrB             = jjnr[jidx+1];
 435             jnrC             = jjnr[jidx+2];
 436             jnrD             = jjnr[jidx+3];
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451
 452             /* Calculate squared distance and things based on it */
 453             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 454
 455             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 456
 457             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq00             = _mm_mul_ps(iq0,jq0);
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 472
 473             fscal            = felec;
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm_mul_ps(fscal,dx00);
 477             ty               = _mm_mul_ps(fscal,dy00);
 478             tz               = _mm_mul_ps(fscal,dz00);
 479
 480             /* Update vectorial force */
 481             fix0             = _mm_add_ps(fix0,tx);
 482             fiy0             = _mm_add_ps(fiy0,ty);
 483             fiz0             = _mm_add_ps(fiz0,tz);
 484
 485             fjptrA             = f+j_coord_offsetA;
 486             fjptrB             = f+j_coord_offsetB;
 487             fjptrC             = f+j_coord_offsetC;
 488             fjptrD             = f+j_coord_offsetD;
 489             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 490
 491             /* Inner loop uses 27 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrlistA         = jjnr[jidx];
 499             jnrlistB         = jjnr[jidx+1];
 500             jnrlistC         = jjnr[jidx+2];
 501             jnrlistD         = jjnr[jidx+3];
 502             /* Sign of each element will be negative for non-real atoms.
 503              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 504              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 505              */
 506             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 507             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 508             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 509             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 510             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 511             j_coord_offsetA  = DIM*jnrA;
 512             j_coord_offsetB  = DIM*jnrB;
 513             j_coord_offsetC  = DIM*jnrC;
 514             j_coord_offsetD  = DIM*jnrD;
 515
 516             /* load j atom coordinates */
 517             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 518                                               x+j_coord_offsetC,x+j_coord_offsetD,
 519                                               &jx0,&jy0,&jz0);
 520
 521             /* Calculate displacement vector */
 522             dx00             = _mm_sub_ps(ix0,jx0);
 523             dy00             = _mm_sub_ps(iy0,jy0);
 524             dz00             = _mm_sub_ps(iz0,jz0);
 525
 526             /* Calculate squared distance and things based on it */
 527             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 528
 529             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 530
 531             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 535                                                               charge+jnrC+0,charge+jnrD+0);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq00             = _mm_mul_ps(iq0,jq0);
 543
 544             /* REACTION-FIELD ELECTROSTATICS */
 545             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm_mul_ps(fscal,dx00);
 553             ty               = _mm_mul_ps(fscal,dy00);
 554             tz               = _mm_mul_ps(fscal,dz00);
 555
 556             /* Update vectorial force */
 557             fix0             = _mm_add_ps(fix0,tx);
 558             fiy0             = _mm_add_ps(fiy0,ty);
 559             fiz0             = _mm_add_ps(fiz0,tz);
 560
 561             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 562             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 563             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 564             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 565             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 566
 567             /* Inner loop uses 27 flops */
 568         }
 569
 570         /* End of innermost loop */
 571
 572         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 573                                               f+i_coord_offset,fshift+i_shift_offset);
 574
 575         /* Increment number of inner iterations */
 576         inneriter                  += j_index_end - j_index_start;
 577
 578         /* Outer loop uses 7 flops */
 579     }
 580
 581     /* Increment number of outer iterations */
 582     outeriter        += nri;
 583
 584     /* Update outer/inner flops */
 585
 586     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 587 }