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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_sse4_1_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 114     rcutoff_scalar   = fr->rcoulomb;
 115     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm_setzero_ps();
 151         fiy0             = _mm_setzero_ps();
 152         fiz0             = _mm_setzero_ps();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm_sub_ps(ix0,jx0);
 181             dy00             = _mm_sub_ps(iy0,jy0);
 182             dz00             = _mm_sub_ps(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 188
 189             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 193                                                               charge+jnrC+0,charge+jnrD+0);
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq00             = _mm_mul_ps(iq0,jq0);
 204
 205             /* REACTION-FIELD ELECTROSTATICS */
 206             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 207             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 208
 209             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velec            = _mm_and_ps(velec,cutoff_mask);
 213             velecsum         = _mm_add_ps(velecsum,velec);
 214
 215             fscal            = felec;
 216
 217             fscal            = _mm_and_ps(fscal,cutoff_mask);
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm_mul_ps(fscal,dx00);
 221             ty               = _mm_mul_ps(fscal,dy00);
 222             tz               = _mm_mul_ps(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_add_ps(fix0,tx);
 226             fiy0             = _mm_add_ps(fiy0,ty);
 227             fiz0             = _mm_add_ps(fiz0,tz);
 228
 229             fjptrA             = f+j_coord_offsetA;
 230             fjptrB             = f+j_coord_offsetB;
 231             fjptrC             = f+j_coord_offsetC;
 232             fjptrD             = f+j_coord_offsetD;
 233             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 234
 235             }
 236
 237             /* Inner loop uses 36 flops */
 238         }
 239
 240         if(jidx<j_index_end)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrlistA         = jjnr[jidx];
 245             jnrlistB         = jjnr[jidx+1];
 246             jnrlistC         = jjnr[jidx+2];
 247             jnrlistD         = jjnr[jidx+3];
 248             /* Sign of each element will be negative for non-real atoms.
 249              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 250              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 251              */
 252             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 253             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 254             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 255             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 256             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 257             j_coord_offsetA  = DIM*jnrA;
 258             j_coord_offsetB  = DIM*jnrB;
 259             j_coord_offsetC  = DIM*jnrC;
 260             j_coord_offsetD  = DIM*jnrD;
 261
 262             /* load j atom coordinates */
 263             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 264                                               x+j_coord_offsetC,x+j_coord_offsetD,
 265                                               &jx0,&jy0,&jz0);
 266
 267             /* Calculate displacement vector */
 268             dx00             = _mm_sub_ps(ix0,jx0);
 269             dy00             = _mm_sub_ps(iy0,jy0);
 270             dz00             = _mm_sub_ps(iz0,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 274
 275             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 276
 277             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 278
 279             /* Load parameters for j particles */
 280             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 281                                                               charge+jnrC+0,charge+jnrD+0);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             if (gmx_mm_any_lt(rsq00,rcutoff2))
 288             {
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq00             = _mm_mul_ps(iq0,jq0);
 292
 293             /* REACTION-FIELD ELECTROSTATICS */
 294             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 295             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 296
 297             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_and_ps(velec,cutoff_mask);
 301             velec            = _mm_andnot_ps(dummy_mask,velec);
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             fscal            = _mm_and_ps(fscal,cutoff_mask);
 307
 308             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm_mul_ps(fscal,dx00);
 312             ty               = _mm_mul_ps(fscal,dy00);
 313             tz               = _mm_mul_ps(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_add_ps(fix0,tx);
 317             fiy0             = _mm_add_ps(fiy0,ty);
 318             fiz0             = _mm_add_ps(fiz0,tz);
 319
 320             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 321             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 322             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 323             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 324             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 325
 326             }
 327
 328             /* Inner loop uses 36 flops */
 329         }
 330
 331         /* End of innermost loop */
 332
 333         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 334                                               f+i_coord_offset,fshift+i_shift_offset);
 335
 336         ggid                        = gid[iidx];
 337         /* Update potential energies */
 338         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 339
 340         /* Increment number of inner iterations */
 341         inneriter                  += j_index_end - j_index_start;
 342
 343         /* Outer loop uses 8 flops */
 344     }
 345
 346     /* Increment number of outer iterations */
 347     outeriter        += nri;
 348
 349     /* Update outer/inner flops */
 350
 351     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 352 }
 353 /*
 354  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 355  * Electrostatics interaction: ReactionField
 356  * VdW interaction:            None
 357  * Geometry:                   Particle-Particle
 358  * Calculate force/pot:        Force
 359  */
 360 void
 361 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 362                     (t_nblist                    * gmx_restrict       nlist,
 363                      rvec                        * gmx_restrict          xx,
 364                      rvec                        * gmx_restrict          ff,
 365                      t_forcerec                  * gmx_restrict          fr,
 366                      t_mdatoms                   * gmx_restrict     mdatoms,
 367                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 368                      t_nrnb                      * gmx_restrict        nrnb)
 369 {
 370     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 371      * just 0 for non-waters.
 372      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 373      * jnr indices corresponding to data put in the four positions in the SIMD register.
 374      */
 375     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 376     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 377     int              jnrA,jnrB,jnrC,jnrD;
 378     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 379     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 381     real             rcutoff_scalar;
 382     real             *shiftvec,*fshift,*x,*f;
 383     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 384     real             scratch[4*DIM];
 385     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 386     int              vdwioffset0;
 387     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 388     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 389     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 390     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 391     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 392     real             *charge;
 393     __m128           dummy_mask,cutoff_mask;
 394     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 395     __m128           one     = _mm_set1_ps(1.0);
 396     __m128           two     = _mm_set1_ps(2.0);
 397     x                = xx[0];
 398     f                = ff[0];
 399
 400     nri              = nlist->nri;
 401     iinr             = nlist->iinr;
 402     jindex           = nlist->jindex;
 403     jjnr             = nlist->jjnr;
 404     shiftidx         = nlist->shift;
 405     gid              = nlist->gid;
 406     shiftvec         = fr->shift_vec[0];
 407     fshift           = fr->fshift[0];
 408     facel            = _mm_set1_ps(fr->epsfac);
 409     charge           = mdatoms->chargeA;
 410     krf              = _mm_set1_ps(fr->ic->k_rf);
 411     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 412     crf              = _mm_set1_ps(fr->ic->c_rf);
 413
 414     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 415     rcutoff_scalar   = fr->rcoulomb;
 416     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 417     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 418
 419     /* Avoid stupid compiler warnings */
 420     jnrA = jnrB = jnrC = jnrD = 0;
 421     j_coord_offsetA = 0;
 422     j_coord_offsetB = 0;
 423     j_coord_offsetC = 0;
 424     j_coord_offsetD = 0;
 425
 426     outeriter        = 0;
 427     inneriter        = 0;
 428
 429     for(iidx=0;iidx<4*DIM;iidx++)
 430     {
 431         scratch[iidx] = 0.0;
 432     }
 433
 434     /* Start outer loop over neighborlists */
 435     for(iidx=0; iidx<nri; iidx++)
 436     {
 437         /* Load shift vector for this list */
 438         i_shift_offset   = DIM*shiftidx[iidx];
 439
 440         /* Load limits for loop over neighbors */
 441         j_index_start    = jindex[iidx];
 442         j_index_end      = jindex[iidx+1];
 443
 444         /* Get outer coordinate index */
 445         inr              = iinr[iidx];
 446         i_coord_offset   = DIM*inr;
 447
 448         /* Load i particle coords and add shift vector */
 449         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 450
 451         fix0             = _mm_setzero_ps();
 452         fiy0             = _mm_setzero_ps();
 453         fiz0             = _mm_setzero_ps();
 454
 455         /* Load parameters for i particles */
 456         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 457
 458         /* Start inner kernel loop */
 459         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrA             = jjnr[jidx];
 464             jnrB             = jjnr[jidx+1];
 465             jnrC             = jjnr[jidx+2];
 466             jnrD             = jjnr[jidx+3];
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469             j_coord_offsetC  = DIM*jnrC;
 470             j_coord_offsetD  = DIM*jnrD;
 471
 472             /* load j atom coordinates */
 473             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               x+j_coord_offsetC,x+j_coord_offsetD,
 475                                               &jx0,&jy0,&jz0);
 476
 477             /* Calculate displacement vector */
 478             dx00             = _mm_sub_ps(ix0,jx0);
 479             dy00             = _mm_sub_ps(iy0,jy0);
 480             dz00             = _mm_sub_ps(iz0,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 484
 485             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 486
 487             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                               charge+jnrC+0,charge+jnrD+0);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             if (gmx_mm_any_lt(rsq00,rcutoff2))
 498             {
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq00             = _mm_mul_ps(iq0,jq0);
 502
 503             /* REACTION-FIELD ELECTROSTATICS */
 504             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 505
 506             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 507
 508             fscal            = felec;
 509
 510             fscal            = _mm_and_ps(fscal,cutoff_mask);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm_mul_ps(fscal,dx00);
 514             ty               = _mm_mul_ps(fscal,dy00);
 515             tz               = _mm_mul_ps(fscal,dz00);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm_add_ps(fix0,tx);
 519             fiy0             = _mm_add_ps(fiy0,ty);
 520             fiz0             = _mm_add_ps(fiz0,tz);
 521
 522             fjptrA             = f+j_coord_offsetA;
 523             fjptrB             = f+j_coord_offsetB;
 524             fjptrC             = f+j_coord_offsetC;
 525             fjptrD             = f+j_coord_offsetD;
 526             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 527
 528             }
 529
 530             /* Inner loop uses 30 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             /* Get j neighbor index, and coordinate index */
 537             jnrlistA         = jjnr[jidx];
 538             jnrlistB         = jjnr[jidx+1];
 539             jnrlistC         = jjnr[jidx+2];
 540             jnrlistD         = jjnr[jidx+3];
 541             /* Sign of each element will be negative for non-real atoms.
 542              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 543              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 544              */
 545             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 546             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 547             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 548             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 549             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 550             j_coord_offsetA  = DIM*jnrA;
 551             j_coord_offsetB  = DIM*jnrB;
 552             j_coord_offsetC  = DIM*jnrC;
 553             j_coord_offsetD  = DIM*jnrD;
 554
 555             /* load j atom coordinates */
 556             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 557                                               x+j_coord_offsetC,x+j_coord_offsetD,
 558                                               &jx0,&jy0,&jz0);
 559
 560             /* Calculate displacement vector */
 561             dx00             = _mm_sub_ps(ix0,jx0);
 562             dy00             = _mm_sub_ps(iy0,jy0);
 563             dz00             = _mm_sub_ps(iz0,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 567
 568             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 569
 570             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 571
 572             /* Load parameters for j particles */
 573             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 574                                                               charge+jnrC+0,charge+jnrD+0);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             if (gmx_mm_any_lt(rsq00,rcutoff2))
 581             {
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq00             = _mm_mul_ps(iq0,jq0);
 585
 586             /* REACTION-FIELD ELECTROSTATICS */
 587             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 588
 589             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_and_ps(fscal,cutoff_mask);
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_ps(fscal,dx00);
 599             ty               = _mm_mul_ps(fscal,dy00);
 600             tz               = _mm_mul_ps(fscal,dz00);
 601
 602             /* Update vectorial force */
 603             fix0             = _mm_add_ps(fix0,tx);
 604             fiy0             = _mm_add_ps(fiy0,ty);
 605             fiz0             = _mm_add_ps(fiz0,tz);
 606
 607             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 608             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 609             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 610             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 611             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 612
 613             }
 614
 615             /* Inner loop uses 30 flops */
 616         }
 617
 618         /* End of innermost loop */
 619
 620         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 621                                               f+i_coord_offset,fshift+i_shift_offset);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 7 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 635 }