src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->ic->rcoulomb;
 112     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinv00           = sse41_invsqrt_f(rsq00);
 185
 186             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 190                                                               charge+jnrC+0,charge+jnrD+0);
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             if (gmx_mm_any_lt(rsq00,rcutoff2))
 197             {
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm_mul_ps(iq0,jq0);
 201
 202             /* REACTION-FIELD ELECTROSTATICS */
 203             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 204             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 205
 206             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             velec            = _mm_and_ps(velec,cutoff_mask);
 210             velecsum         = _mm_add_ps(velecsum,velec);
 211
 212             fscal            = felec;
 213
 214             fscal            = _mm_and_ps(fscal,cutoff_mask);
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_ps(fscal,dx00);
 218             ty               = _mm_mul_ps(fscal,dy00);
 219             tz               = _mm_mul_ps(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_ps(fix0,tx);
 223             fiy0             = _mm_add_ps(fiy0,ty);
 224             fiz0             = _mm_add_ps(fiz0,tz);
 225
 226             fjptrA             = f+j_coord_offsetA;
 227             fjptrB             = f+j_coord_offsetB;
 228             fjptrC             = f+j_coord_offsetC;
 229             fjptrD             = f+j_coord_offsetD;
 230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 231
 232             }
 233
 234             /* Inner loop uses 36 flops */
 235         }
 236
 237         if(jidx<j_index_end)
 238         {
 239
 240             /* Get j neighbor index, and coordinate index */
 241             jnrlistA         = jjnr[jidx];
 242             jnrlistB         = jjnr[jidx+1];
 243             jnrlistC         = jjnr[jidx+2];
 244             jnrlistD         = jjnr[jidx+3];
 245             /* Sign of each element will be negative for non-real atoms.
 246              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 247              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 248              */
 249             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 250             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 251             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 252             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 253             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 254             j_coord_offsetA  = DIM*jnrA;
 255             j_coord_offsetB  = DIM*jnrB;
 256             j_coord_offsetC  = DIM*jnrC;
 257             j_coord_offsetD  = DIM*jnrD;
 258
 259             /* load j atom coordinates */
 260             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 261                                               x+j_coord_offsetC,x+j_coord_offsetD,
 262                                               &jx0,&jy0,&jz0);
 263
 264             /* Calculate displacement vector */
 265             dx00             = _mm_sub_ps(ix0,jx0);
 266             dy00             = _mm_sub_ps(iy0,jy0);
 267             dz00             = _mm_sub_ps(iz0,jz0);
 268
 269             /* Calculate squared distance and things based on it */
 270             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 271
 272             rinv00           = sse41_invsqrt_f(rsq00);
 273
 274             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 275
 276             /* Load parameters for j particles */
 277             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 278                                                               charge+jnrC+0,charge+jnrD+0);
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             if (gmx_mm_any_lt(rsq00,rcutoff2))
 285             {
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq00             = _mm_mul_ps(iq0,jq0);
 289
 290             /* REACTION-FIELD ELECTROSTATICS */
 291             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 292             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 293
 294             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_ps(velec,cutoff_mask);
 298             velec            = _mm_andnot_ps(dummy_mask,velec);
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_and_ps(fscal,cutoff_mask);
 304
 305             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_ps(fscal,dx00);
 309             ty               = _mm_mul_ps(fscal,dy00);
 310             tz               = _mm_mul_ps(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_add_ps(fix0,tx);
 314             fiy0             = _mm_add_ps(fiy0,ty);
 315             fiz0             = _mm_add_ps(fiz0,tz);
 316
 317             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 318             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 319             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 320             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 321             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 322
 323             }
 324
 325             /* Inner loop uses 36 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 331                                               f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 336
 337         /* Increment number of inner iterations */
 338         inneriter                  += j_index_end - j_index_start;
 339
 340         /* Outer loop uses 8 flops */
 341     }
 342
 343     /* Increment number of outer iterations */
 344     outeriter        += nri;
 345
 346     /* Update outer/inner flops */
 347
 348     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 349 }
 350 /*
 351  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 352  * Electrostatics interaction: ReactionField
 353  * VdW interaction:            None
 354  * Geometry:                   Particle-Particle
 355  * Calculate force/pot:        Force
 356  */
 357 void
 358 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 359                     (t_nblist                    * gmx_restrict       nlist,
 360                      rvec                        * gmx_restrict          xx,
 361                      rvec                        * gmx_restrict          ff,
 362                      struct t_forcerec           * gmx_restrict          fr,
 363                      t_mdatoms                   * gmx_restrict     mdatoms,
 364                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 365                      t_nrnb                      * gmx_restrict        nrnb)
 366 {
 367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 368      * just 0 for non-waters.
 369      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 370      * jnr indices corresponding to data put in the four positions in the SIMD register.
 371      */
 372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 374     int              jnrA,jnrB,jnrC,jnrD;
 375     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 376     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 377     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 378     real             rcutoff_scalar;
 379     real             *shiftvec,*fshift,*x,*f;
 380     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 381     real             scratch[4*DIM];
 382     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 383     int              vdwioffset0;
 384     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 385     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 386     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 387     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 388     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 389     real             *charge;
 390     __m128           dummy_mask,cutoff_mask;
 391     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 392     __m128           one     = _mm_set1_ps(1.0);
 393     __m128           two     = _mm_set1_ps(2.0);
 394     x                = xx[0];
 395     f                = ff[0];
 396
 397     nri              = nlist->nri;
 398     iinr             = nlist->iinr;
 399     jindex           = nlist->jindex;
 400     jjnr             = nlist->jjnr;
 401     shiftidx         = nlist->shift;
 402     gid              = nlist->gid;
 403     shiftvec         = fr->shift_vec[0];
 404     fshift           = fr->fshift[0];
 405     facel            = _mm_set1_ps(fr->ic->epsfac);
 406     charge           = mdatoms->chargeA;
 407     krf              = _mm_set1_ps(fr->ic->k_rf);
 408     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 409     crf              = _mm_set1_ps(fr->ic->c_rf);
 410
 411     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 412     rcutoff_scalar   = fr->ic->rcoulomb;
 413     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 414     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = jnrC = jnrD = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420     j_coord_offsetC = 0;
 421     j_coord_offsetD = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     for(iidx=0;iidx<4*DIM;iidx++)
 427     {
 428         scratch[iidx] = 0.0;
 429     }
 430
 431     /* Start outer loop over neighborlists */
 432     for(iidx=0; iidx<nri; iidx++)
 433     {
 434         /* Load shift vector for this list */
 435         i_shift_offset   = DIM*shiftidx[iidx];
 436
 437         /* Load limits for loop over neighbors */
 438         j_index_start    = jindex[iidx];
 439         j_index_end      = jindex[iidx+1];
 440
 441         /* Get outer coordinate index */
 442         inr              = iinr[iidx];
 443         i_coord_offset   = DIM*inr;
 444
 445         /* Load i particle coords and add shift vector */
 446         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 447
 448         fix0             = _mm_setzero_ps();
 449         fiy0             = _mm_setzero_ps();
 450         fiz0             = _mm_setzero_ps();
 451
 452         /* Load parameters for i particles */
 453         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 454
 455         /* Start inner kernel loop */
 456         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrA             = jjnr[jidx];
 461             jnrB             = jjnr[jidx+1];
 462             jnrC             = jjnr[jidx+2];
 463             jnrD             = jjnr[jidx+3];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466             j_coord_offsetC  = DIM*jnrC;
 467             j_coord_offsetD  = DIM*jnrD;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                               x+j_coord_offsetC,x+j_coord_offsetD,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm_sub_ps(ix0,jx0);
 476             dy00             = _mm_sub_ps(iy0,jy0);
 477             dz00             = _mm_sub_ps(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481
 482             rinv00           = sse41_invsqrt_f(rsq00);
 483
 484             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                               charge+jnrC+0,charge+jnrD+0);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm_any_lt(rsq00,rcutoff2))
 495             {
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq00             = _mm_mul_ps(iq0,jq0);
 499
 500             /* REACTION-FIELD ELECTROSTATICS */
 501             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 502
 503             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_and_ps(fscal,cutoff_mask);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx00);
 511             ty               = _mm_mul_ps(fscal,dy00);
 512             tz               = _mm_mul_ps(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_ps(fix0,tx);
 516             fiy0             = _mm_add_ps(fiy0,ty);
 517             fiz0             = _mm_add_ps(fiz0,tz);
 518
 519             fjptrA             = f+j_coord_offsetA;
 520             fjptrB             = f+j_coord_offsetB;
 521             fjptrC             = f+j_coord_offsetC;
 522             fjptrD             = f+j_coord_offsetD;
 523             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 524
 525             }
 526
 527             /* Inner loop uses 30 flops */
 528         }
 529
 530         if(jidx<j_index_end)
 531         {
 532
 533             /* Get j neighbor index, and coordinate index */
 534             jnrlistA         = jjnr[jidx];
 535             jnrlistB         = jjnr[jidx+1];
 536             jnrlistC         = jjnr[jidx+2];
 537             jnrlistD         = jjnr[jidx+3];
 538             /* Sign of each element will be negative for non-real atoms.
 539              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 540              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 541              */
 542             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 543             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 544             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 545             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 546             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 547             j_coord_offsetA  = DIM*jnrA;
 548             j_coord_offsetB  = DIM*jnrB;
 549             j_coord_offsetC  = DIM*jnrC;
 550             j_coord_offsetD  = DIM*jnrD;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 554                                               x+j_coord_offsetC,x+j_coord_offsetD,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_ps(ix0,jx0);
 559             dy00             = _mm_sub_ps(iy0,jy0);
 560             dz00             = _mm_sub_ps(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 564
 565             rinv00           = sse41_invsqrt_f(rsq00);
 566
 567             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 571                                                               charge+jnrC+0,charge+jnrD+0);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq00,rcutoff2))
 578             {
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq00             = _mm_mul_ps(iq0,jq0);
 582
 583             /* REACTION-FIELD ELECTROSTATICS */
 584             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 585
 586             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_and_ps(fscal,cutoff_mask);
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx00);
 596             ty               = _mm_mul_ps(fscal,dy00);
 597             tz               = _mm_mul_ps(fscal,dz00);
 598
 599             /* Update vectorial force */
 600             fix0             = _mm_add_ps(fix0,tx);
 601             fiy0             = _mm_add_ps(fiy0,ty);
 602             fiz0             = _mm_add_ps(fiz0,tz);
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 609
 610             }
 611
 612             /* Inner loop uses 30 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 618                                               f+i_coord_offset,fshift+i_shift_offset);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 7 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 632 }