src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_avx_128_fma_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 avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm_set1_ps(fr->ic->c_rf);
 110
 111     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 112     rcutoff_scalar   = fr->rcoulomb;
 113     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 114     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_ps();
 149         fiy0             = _mm_setzero_ps();
 150         fiz0             = _mm_setzero_ps();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_ps();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             jnrC             = jjnr[jidx+2];
 166             jnrD             = jjnr[jidx+3];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               x+j_coord_offsetC,x+j_coord_offsetD,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_ps(ix0,jx0);
 179             dy00             = _mm_sub_ps(iy0,jy0);
 180             dz00             = _mm_sub_ps(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 186
 187             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 191                                                               charge+jnrC+0,charge+jnrD+0);
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             if (gmx_mm_any_lt(rsq00,rcutoff2))
 198             {
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_ps(iq0,jq0);
 202
 203             /* REACTION-FIELD ELECTROSTATICS */
 204             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 205             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 206
 207             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 208
 209             /* Update potential sum for this i atom from the interaction with this j atom. */
 210             velec            = _mm_and_ps(velec,cutoff_mask);
 211             velecsum         = _mm_add_ps(velecsum,velec);
 212
 213             fscal            = felec;
 214
 215             fscal            = _mm_and_ps(fscal,cutoff_mask);
 216
 217              /* Update vectorial force */
 218             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 219             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 220             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 221
 222             fjptrA             = f+j_coord_offsetA;
 223             fjptrB             = f+j_coord_offsetB;
 224             fjptrC             = f+j_coord_offsetC;
 225             fjptrD             = f+j_coord_offsetD;
 226             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 227                                                    _mm_mul_ps(dx00,fscal),
 228                                                    _mm_mul_ps(dy00,fscal),
 229                                                    _mm_mul_ps(dz00,fscal));
 230
 231             }
 232
 233             /* Inner loop uses 39 flops */
 234         }
 235
 236         if(jidx<j_index_end)
 237         {
 238
 239             /* Get j neighbor index, and coordinate index */
 240             jnrlistA         = jjnr[jidx];
 241             jnrlistB         = jjnr[jidx+1];
 242             jnrlistC         = jjnr[jidx+2];
 243             jnrlistD         = jjnr[jidx+3];
 244             /* Sign of each element will be negative for non-real atoms.
 245              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 246              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 247              */
 248             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 249             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 250             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 251             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 252             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 253             j_coord_offsetA  = DIM*jnrA;
 254             j_coord_offsetB  = DIM*jnrB;
 255             j_coord_offsetC  = DIM*jnrC;
 256             j_coord_offsetD  = DIM*jnrD;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 260                                               x+j_coord_offsetC,x+j_coord_offsetD,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_ps(ix0,jx0);
 265             dy00             = _mm_sub_ps(iy0,jy0);
 266             dz00             = _mm_sub_ps(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 272
 273             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 277                                                               charge+jnrC+0,charge+jnrD+0);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             if (gmx_mm_any_lt(rsq00,rcutoff2))
 284             {
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq00             = _mm_mul_ps(iq0,jq0);
 288
 289             /* REACTION-FIELD ELECTROSTATICS */
 290             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 291             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 292
 293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_ps(velec,cutoff_mask);
 297             velec            = _mm_andnot_ps(dummy_mask,velec);
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             fscal            = _mm_and_ps(fscal,cutoff_mask);
 303
 304             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 312             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 313             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 314             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 315             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 316                                                    _mm_mul_ps(dx00,fscal),
 317                                                    _mm_mul_ps(dy00,fscal),
 318                                                    _mm_mul_ps(dz00,fscal));
 319
 320             }
 321
 322             /* Inner loop uses 39 flops */
 323         }
 324
 325         /* End of innermost loop */
 326
 327         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 328                                               f+i_coord_offset,fshift+i_shift_offset);
 329
 330         ggid                        = gid[iidx];
 331         /* Update potential energies */
 332         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 333
 334         /* Increment number of inner iterations */
 335         inneriter                  += j_index_end - j_index_start;
 336
 337         /* Outer loop uses 8 flops */
 338     }
 339
 340     /* Increment number of outer iterations */
 341     outeriter        += nri;
 342
 343     /* Update outer/inner flops */
 344
 345     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*39);
 346 }
 347 /*
 348  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_single
 349  * Electrostatics interaction: ReactionField
 350  * VdW interaction:            None
 351  * Geometry:                   Particle-Particle
 352  * Calculate force/pot:        Force
 353  */
 354 void
 355 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_single
 356                     (t_nblist                    * gmx_restrict       nlist,
 357                      rvec                        * gmx_restrict          xx,
 358                      rvec                        * gmx_restrict          ff,
 359                      t_forcerec                  * gmx_restrict          fr,
 360                      t_mdatoms                   * gmx_restrict     mdatoms,
 361                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 362                      t_nrnb                      * gmx_restrict        nrnb)
 363 {
 364     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 365      * just 0 for non-waters.
 366      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 367      * jnr indices corresponding to data put in the four positions in the SIMD register.
 368      */
 369     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 370     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 371     int              jnrA,jnrB,jnrC,jnrD;
 372     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 373     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 374     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 375     real             rcutoff_scalar;
 376     real             *shiftvec,*fshift,*x,*f;
 377     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 378     real             scratch[4*DIM];
 379     __m128           fscal,rcutoff,rcutoff2,jidxall;
 380     int              vdwioffset0;
 381     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 382     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 383     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 384     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 385     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 386     real             *charge;
 387     __m128           dummy_mask,cutoff_mask;
 388     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 389     __m128           one     = _mm_set1_ps(1.0);
 390     __m128           two     = _mm_set1_ps(2.0);
 391     x                = xx[0];
 392     f                = ff[0];
 393
 394     nri              = nlist->nri;
 395     iinr             = nlist->iinr;
 396     jindex           = nlist->jindex;
 397     jjnr             = nlist->jjnr;
 398     shiftidx         = nlist->shift;
 399     gid              = nlist->gid;
 400     shiftvec         = fr->shift_vec[0];
 401     fshift           = fr->fshift[0];
 402     facel            = _mm_set1_ps(fr->epsfac);
 403     charge           = mdatoms->chargeA;
 404     krf              = _mm_set1_ps(fr->ic->k_rf);
 405     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 406     crf              = _mm_set1_ps(fr->ic->c_rf);
 407
 408     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 409     rcutoff_scalar   = fr->rcoulomb;
 410     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 411     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 412
 413     /* Avoid stupid compiler warnings */
 414     jnrA = jnrB = jnrC = jnrD = 0;
 415     j_coord_offsetA = 0;
 416     j_coord_offsetB = 0;
 417     j_coord_offsetC = 0;
 418     j_coord_offsetD = 0;
 419
 420     outeriter        = 0;
 421     inneriter        = 0;
 422
 423     for(iidx=0;iidx<4*DIM;iidx++)
 424     {
 425         scratch[iidx] = 0.0;
 426     }
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433
 434         /* Load limits for loop over neighbors */
 435         j_index_start    = jindex[iidx];
 436         j_index_end      = jindex[iidx+1];
 437
 438         /* Get outer coordinate index */
 439         inr              = iinr[iidx];
 440         i_coord_offset   = DIM*inr;
 441
 442         /* Load i particle coords and add shift vector */
 443         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 444
 445         fix0             = _mm_setzero_ps();
 446         fiy0             = _mm_setzero_ps();
 447         fiz0             = _mm_setzero_ps();
 448
 449         /* Load parameters for i particles */
 450         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 451
 452         /* Start inner kernel loop */
 453         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrA             = jjnr[jidx];
 458             jnrB             = jjnr[jidx+1];
 459             jnrC             = jjnr[jidx+2];
 460             jnrD             = jjnr[jidx+3];
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463             j_coord_offsetC  = DIM*jnrC;
 464             j_coord_offsetD  = DIM*jnrD;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                               x+j_coord_offsetC,x+j_coord_offsetD,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm_sub_ps(ix0,jx0);
 473             dy00             = _mm_sub_ps(iy0,jy0);
 474             dz00             = _mm_sub_ps(iz0,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 478
 479             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 480
 481             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 485                                                               charge+jnrC+0,charge+jnrD+0);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq00,rcutoff2))
 492             {
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm_mul_ps(iq0,jq0);
 496
 497             /* REACTION-FIELD ELECTROSTATICS */
 498             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 499
 500             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm_and_ps(fscal,cutoff_mask);
 505
 506              /* Update vectorial force */
 507             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 508             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 509             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 510
 511             fjptrA             = f+j_coord_offsetA;
 512             fjptrB             = f+j_coord_offsetB;
 513             fjptrC             = f+j_coord_offsetC;
 514             fjptrD             = f+j_coord_offsetD;
 515             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 516                                                    _mm_mul_ps(dx00,fscal),
 517                                                    _mm_mul_ps(dy00,fscal),
 518                                                    _mm_mul_ps(dz00,fscal));
 519
 520             }
 521
 522             /* Inner loop uses 33 flops */
 523         }
 524
 525         if(jidx<j_index_end)
 526         {
 527
 528             /* Get j neighbor index, and coordinate index */
 529             jnrlistA         = jjnr[jidx];
 530             jnrlistB         = jjnr[jidx+1];
 531             jnrlistC         = jjnr[jidx+2];
 532             jnrlistD         = jjnr[jidx+3];
 533             /* Sign of each element will be negative for non-real atoms.
 534              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 535              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 536              */
 537             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 538             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 539             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 540             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 541             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544             j_coord_offsetC  = DIM*jnrC;
 545             j_coord_offsetD  = DIM*jnrD;
 546
 547             /* load j atom coordinates */
 548             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                               x+j_coord_offsetC,x+j_coord_offsetD,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm_sub_ps(ix0,jx0);
 554             dy00             = _mm_sub_ps(iy0,jy0);
 555             dz00             = _mm_sub_ps(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 561
 562             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 566                                                               charge+jnrC+0,charge+jnrD+0);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq00,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq00             = _mm_mul_ps(iq0,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 580
 581             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm_and_ps(fscal,cutoff_mask);
 586
 587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 588
 589              /* Update vectorial force */
 590             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 591             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 592             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 593
 594             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 595             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 596             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 597             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 598             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 599                                                    _mm_mul_ps(dx00,fscal),
 600                                                    _mm_mul_ps(dy00,fscal),
 601                                                    _mm_mul_ps(dz00,fscal));
 602
 603             }
 604
 605             /* Inner loop uses 33 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 611                                               f+i_coord_offset,fshift+i_shift_offset);
 612
 613         /* Increment number of inner iterations */
 614         inneriter                  += j_index_end - j_index_start;
 615
 616         /* Outer loop uses 7 flops */
 617     }
 618
 619     /* Increment number of outer iterations */
 620     outeriter        += nri;
 621
 622     /* Update outer/inner flops */
 623
 624     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*33);
 625 }