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) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128           dummy_mask,cutoff_mask;
  90     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  91     __m128           one     = _mm_set1_ps(1.0);
  92     __m128           two     = _mm_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_ps(fr->ic->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_ps(fr->ic->k_rf);
 107     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_ps(fr->ic->c_rf);
 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_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_ps();
 143         fiy0             = _mm_setzero_ps();
 144         fiz0             = _mm_setzero_ps();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_ps();
 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_mm_load_1rvec_4ptr_swizzle_ps(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             = _mm_sub_ps(ix0,jx0);
 173             dy00             = _mm_sub_ps(iy0,jy0);
 174             dz00             = _mm_sub_ps(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 178
 179             rinv00           = sse41_invsqrt_f(rsq00);
 180
 181             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_4real_swizzle_ps(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             = _mm_mul_ps(iq0,jq0);
 193
 194             /* REACTION-FIELD ELECTROSTATICS */
 195             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 196             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 197
 198             /* Update potential sum for this i atom from the interaction with this j atom. */
 199             velecsum         = _mm_add_ps(velecsum,velec);
 200
 201             fscal            = felec;
 202
 203             /* Calculate temporary vectorial force */
 204             tx               = _mm_mul_ps(fscal,dx00);
 205             ty               = _mm_mul_ps(fscal,dy00);
 206             tz               = _mm_mul_ps(fscal,dz00);
 207
 208             /* Update vectorial force */
 209             fix0             = _mm_add_ps(fix0,tx);
 210             fiy0             = _mm_add_ps(fiy0,ty);
 211             fiz0             = _mm_add_ps(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_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 218
 219             /* Inner loop uses 32 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_ps(mask,val) to clear dummy entries.
 233              */
 234             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 235             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 236             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 237             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 238             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 239             j_coord_offsetA  = DIM*jnrA;
 240             j_coord_offsetB  = DIM*jnrB;
 241             j_coord_offsetC  = DIM*jnrC;
 242             j_coord_offsetD  = DIM*jnrD;
 243
 244             /* load j atom coordinates */
 245             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 246                                               x+j_coord_offsetC,x+j_coord_offsetD,
 247                                               &jx0,&jy0,&jz0);
 248
 249             /* Calculate displacement vector */
 250             dx00             = _mm_sub_ps(ix0,jx0);
 251             dy00             = _mm_sub_ps(iy0,jy0);
 252             dz00             = _mm_sub_ps(iz0,jz0);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 256
 257             rinv00           = sse41_invsqrt_f(rsq00);
 258
 259             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                               charge+jnrC+0,charge+jnrD+0);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq00             = _mm_mul_ps(iq0,jq0);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 274             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velec            = _mm_andnot_ps(dummy_mask,velec);
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_ps(fscal,dx00);
 286             ty               = _mm_mul_ps(fscal,dy00);
 287             tz               = _mm_mul_ps(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_ps(fix0,tx);
 291             fiy0             = _mm_add_ps(fiy0,ty);
 292             fiz0             = _mm_add_ps(fiz0,tz);
 293
 294             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 295             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 296             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 297             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 299
 300             /* Inner loop uses 32 flops */
 301         }
 302
 303         /* End of innermost loop */
 304
 305         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 306                                               f+i_coord_offset,fshift+i_shift_offset);
 307
 308         ggid                        = gid[iidx];
 309         /* Update potential energies */
 310         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 311
 312         /* Increment number of inner iterations */
 313         inneriter                  += j_index_end - j_index_start;
 314
 315         /* Outer loop uses 8 flops */
 316     }
 317
 318     /* Increment number of outer iterations */
 319     outeriter        += nri;
 320
 321     /* Update outer/inner flops */
 322
 323     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 324 }
 325 /*
 326  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse4_1_single
 327  * Electrostatics interaction: ReactionField
 328  * VdW interaction:            None
 329  * Geometry:                   Particle-Particle
 330  * Calculate force/pot:        Force
 331  */
 332 void
 333 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse4_1_single
 334                     (t_nblist                    * gmx_restrict       nlist,
 335                      rvec                        * gmx_restrict          xx,
 336                      rvec                        * gmx_restrict          ff,
 337                      struct t_forcerec           * gmx_restrict          fr,
 338                      t_mdatoms                   * gmx_restrict     mdatoms,
 339                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 340                      t_nrnb                      * gmx_restrict        nrnb)
 341 {
 342     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 343      * just 0 for non-waters.
 344      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 345      * jnr indices corresponding to data put in the four positions in the SIMD register.
 346      */
 347     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 348     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 349     int              jnrA,jnrB,jnrC,jnrD;
 350     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 351     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 352     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 353     real             rcutoff_scalar;
 354     real             *shiftvec,*fshift,*x,*f;
 355     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 356     real             scratch[4*DIM];
 357     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 358     int              vdwioffset0;
 359     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 360     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 361     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 362     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 363     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 364     real             *charge;
 365     __m128           dummy_mask,cutoff_mask;
 366     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 367     __m128           one     = _mm_set1_ps(1.0);
 368     __m128           two     = _mm_set1_ps(2.0);
 369     x                = xx[0];
 370     f                = ff[0];
 371
 372     nri              = nlist->nri;
 373     iinr             = nlist->iinr;
 374     jindex           = nlist->jindex;
 375     jjnr             = nlist->jjnr;
 376     shiftidx         = nlist->shift;
 377     gid              = nlist->gid;
 378     shiftvec         = fr->shift_vec[0];
 379     fshift           = fr->fshift[0];
 380     facel            = _mm_set1_ps(fr->ic->epsfac);
 381     charge           = mdatoms->chargeA;
 382     krf              = _mm_set1_ps(fr->ic->k_rf);
 383     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 384     crf              = _mm_set1_ps(fr->ic->c_rf);
 385
 386     /* Avoid stupid compiler warnings */
 387     jnrA = jnrB = jnrC = jnrD = 0;
 388     j_coord_offsetA = 0;
 389     j_coord_offsetB = 0;
 390     j_coord_offsetC = 0;
 391     j_coord_offsetD = 0;
 392
 393     outeriter        = 0;
 394     inneriter        = 0;
 395
 396     for(iidx=0;iidx<4*DIM;iidx++)
 397     {
 398         scratch[iidx] = 0.0;
 399     }
 400
 401     /* Start outer loop over neighborlists */
 402     for(iidx=0; iidx<nri; iidx++)
 403     {
 404         /* Load shift vector for this list */
 405         i_shift_offset   = DIM*shiftidx[iidx];
 406
 407         /* Load limits for loop over neighbors */
 408         j_index_start    = jindex[iidx];
 409         j_index_end      = jindex[iidx+1];
 410
 411         /* Get outer coordinate index */
 412         inr              = iinr[iidx];
 413         i_coord_offset   = DIM*inr;
 414
 415         /* Load i particle coords and add shift vector */
 416         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 417
 418         fix0             = _mm_setzero_ps();
 419         fiy0             = _mm_setzero_ps();
 420         fiz0             = _mm_setzero_ps();
 421
 422         /* Load parameters for i particles */
 423         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 424
 425         /* Start inner kernel loop */
 426         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrA             = jjnr[jidx];
 431             jnrB             = jjnr[jidx+1];
 432             jnrC             = jjnr[jidx+2];
 433             jnrD             = jjnr[jidx+3];
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436             j_coord_offsetC  = DIM*jnrC;
 437             j_coord_offsetD  = DIM*jnrD;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               x+j_coord_offsetC,x+j_coord_offsetD,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_ps(ix0,jx0);
 446             dy00             = _mm_sub_ps(iy0,jy0);
 447             dz00             = _mm_sub_ps(iz0,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 451
 452             rinv00           = sse41_invsqrt_f(rsq00);
 453
 454             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 458                                                               charge+jnrC+0,charge+jnrD+0);
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             /* Compute parameters for interactions between i and j atoms */
 465             qq00             = _mm_mul_ps(iq0,jq0);
 466
 467             /* REACTION-FIELD ELECTROSTATICS */
 468             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 469
 470             fscal            = felec;
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm_mul_ps(fscal,dx00);
 474             ty               = _mm_mul_ps(fscal,dy00);
 475             tz               = _mm_mul_ps(fscal,dz00);
 476
 477             /* Update vectorial force */
 478             fix0             = _mm_add_ps(fix0,tx);
 479             fiy0             = _mm_add_ps(fiy0,ty);
 480             fiz0             = _mm_add_ps(fiz0,tz);
 481
 482             fjptrA             = f+j_coord_offsetA;
 483             fjptrB             = f+j_coord_offsetB;
 484             fjptrC             = f+j_coord_offsetC;
 485             fjptrD             = f+j_coord_offsetD;
 486             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 487
 488             /* Inner loop uses 27 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             /* Get j neighbor index, and coordinate index */
 495             jnrlistA         = jjnr[jidx];
 496             jnrlistB         = jjnr[jidx+1];
 497             jnrlistC         = jjnr[jidx+2];
 498             jnrlistD         = jjnr[jidx+3];
 499             /* Sign of each element will be negative for non-real atoms.
 500              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 501              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 502              */
 503             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 504             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 505             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 506             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 507             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510             j_coord_offsetC  = DIM*jnrC;
 511             j_coord_offsetD  = DIM*jnrD;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               x+j_coord_offsetC,x+j_coord_offsetD,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _mm_sub_ps(ix0,jx0);
 520             dy00             = _mm_sub_ps(iy0,jy0);
 521             dz00             = _mm_sub_ps(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 525
 526             rinv00           = sse41_invsqrt_f(rsq00);
 527
 528             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 532                                                               charge+jnrC+0,charge+jnrD+0);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq00             = _mm_mul_ps(iq0,jq0);
 540
 541             /* REACTION-FIELD ELECTROSTATICS */
 542             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_ps(fscal,dx00);
 550             ty               = _mm_mul_ps(fscal,dy00);
 551             tz               = _mm_mul_ps(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm_add_ps(fix0,tx);
 555             fiy0             = _mm_add_ps(fiy0,ty);
 556             fiz0             = _mm_add_ps(fiz0,tz);
 557
 558             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 559             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 560             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 561             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 562             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 563
 564             /* Inner loop uses 27 flops */
 565         }
 566
 567         /* End of innermost loop */
 568
 569         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 570                                               f+i_coord_offset,fshift+i_shift_offset);
 571
 572         /* Increment number of inner iterations */
 573         inneriter                  += j_index_end - j_index_start;
 574
 575         /* Outer loop uses 7 flops */
 576     }
 577
 578     /* Increment number of outer iterations */
 579     outeriter        += nri;
 580
 581     /* Update outer/inner flops */
 582
 583     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 584 }