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 "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_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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_macc_ps(krf,rsq00,rinv00),crf));
 207             felec            = _mm_mul_ps(qq00,_mm_msub_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              /* Update vectorial force */
 220             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 221             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 222             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 223
 224             fjptrA             = f+j_coord_offsetA;
 225             fjptrB             = f+j_coord_offsetB;
 226             fjptrC             = f+j_coord_offsetC;
 227             fjptrD             = f+j_coord_offsetD;
 228             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 229                                                    _mm_mul_ps(dx00,fscal),
 230                                                    _mm_mul_ps(dy00,fscal),
 231                                                    _mm_mul_ps(dz00,fscal));
 232
 233             }
 234
 235             /* Inner loop uses 39 flops */
 236         }
 237
 238         if(jidx<j_index_end)
 239         {
 240
 241             /* Get j neighbor index, and coordinate index */
 242             jnrlistA         = jjnr[jidx];
 243             jnrlistB         = jjnr[jidx+1];
 244             jnrlistC         = jjnr[jidx+2];
 245             jnrlistD         = jjnr[jidx+3];
 246             /* Sign of each element will be negative for non-real atoms.
 247              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 248              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 249              */
 250             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 251             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 252             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 253             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 254             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 255             j_coord_offsetA  = DIM*jnrA;
 256             j_coord_offsetB  = DIM*jnrB;
 257             j_coord_offsetC  = DIM*jnrC;
 258             j_coord_offsetD  = DIM*jnrD;
 259
 260             /* load j atom coordinates */
 261             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 262                                               x+j_coord_offsetC,x+j_coord_offsetD,
 263                                               &jx0,&jy0,&jz0);
 264
 265             /* Calculate displacement vector */
 266             dx00             = _mm_sub_ps(ix0,jx0);
 267             dy00             = _mm_sub_ps(iy0,jy0);
 268             dz00             = _mm_sub_ps(iz0,jz0);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 272
 273             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 274
 275             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 276
 277             /* Load parameters for j particles */
 278             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 279                                                               charge+jnrC+0,charge+jnrD+0);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq00,rcutoff2))
 286             {
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq00             = _mm_mul_ps(iq0,jq0);
 290
 291             /* REACTION-FIELD ELECTROSTATICS */
 292             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 293             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 294
 295             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_and_ps(velec,cutoff_mask);
 299             velec            = _mm_andnot_ps(dummy_mask,velec);
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             fscal            = _mm_and_ps(fscal,cutoff_mask);
 305
 306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 307
 308              /* Update vectorial force */
 309             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 312
 313             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 314             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 315             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 316             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 317             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 318                                                    _mm_mul_ps(dx00,fscal),
 319                                                    _mm_mul_ps(dy00,fscal),
 320                                                    _mm_mul_ps(dz00,fscal));
 321
 322             }
 323
 324             /* Inner loop uses 39 flops */
 325         }
 326
 327         /* End of innermost loop */
 328
 329         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 330                                               f+i_coord_offset,fshift+i_shift_offset);
 331
 332         ggid                        = gid[iidx];
 333         /* Update potential energies */
 334         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 335
 336         /* Increment number of inner iterations */
 337         inneriter                  += j_index_end - j_index_start;
 338
 339         /* Outer loop uses 8 flops */
 340     }
 341
 342     /* Increment number of outer iterations */
 343     outeriter        += nri;
 344
 345     /* Update outer/inner flops */
 346
 347     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*39);
 348 }
 349 /*
 350  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_single
 351  * Electrostatics interaction: ReactionField
 352  * VdW interaction:            None
 353  * Geometry:                   Particle-Particle
 354  * Calculate force/pot:        Force
 355  */
 356 void
 357 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_single
 358                     (t_nblist                    * gmx_restrict       nlist,
 359                      rvec                        * gmx_restrict          xx,
 360                      rvec                        * gmx_restrict          ff,
 361                      t_forcerec                  * gmx_restrict          fr,
 362                      t_mdatoms                   * gmx_restrict     mdatoms,
 363                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 364                      t_nrnb                      * gmx_restrict        nrnb)
 365 {
 366     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 367      * just 0 for non-waters.
 368      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 369      * jnr indices corresponding to data put in the four positions in the SIMD register.
 370      */
 371     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 372     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 373     int              jnrA,jnrB,jnrC,jnrD;
 374     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 375     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 376     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 377     real             rcutoff_scalar;
 378     real             *shiftvec,*fshift,*x,*f;
 379     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 380     real             scratch[4*DIM];
 381     __m128           fscal,rcutoff,rcutoff2,jidxall;
 382     int              vdwioffset0;
 383     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 384     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 385     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 386     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 387     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 388     real             *charge;
 389     __m128           dummy_mask,cutoff_mask;
 390     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 391     __m128           one     = _mm_set1_ps(1.0);
 392     __m128           two     = _mm_set1_ps(2.0);
 393     x                = xx[0];
 394     f                = ff[0];
 395
 396     nri              = nlist->nri;
 397     iinr             = nlist->iinr;
 398     jindex           = nlist->jindex;
 399     jjnr             = nlist->jjnr;
 400     shiftidx         = nlist->shift;
 401     gid              = nlist->gid;
 402     shiftvec         = fr->shift_vec[0];
 403     fshift           = fr->fshift[0];
 404     facel            = _mm_set1_ps(fr->epsfac);
 405     charge           = mdatoms->chargeA;
 406     krf              = _mm_set1_ps(fr->ic->k_rf);
 407     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 408     crf              = _mm_set1_ps(fr->ic->c_rf);
 409
 410     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 411     rcutoff_scalar   = fr->rcoulomb;
 412     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 413     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 414
 415     /* Avoid stupid compiler warnings */
 416     jnrA = jnrB = jnrC = jnrD = 0;
 417     j_coord_offsetA = 0;
 418     j_coord_offsetB = 0;
 419     j_coord_offsetC = 0;
 420     j_coord_offsetD = 0;
 421
 422     outeriter        = 0;
 423     inneriter        = 0;
 424
 425     for(iidx=0;iidx<4*DIM;iidx++)
 426     {
 427         scratch[iidx] = 0.0;
 428     }
 429
 430     /* Start outer loop over neighborlists */
 431     for(iidx=0; iidx<nri; iidx++)
 432     {
 433         /* Load shift vector for this list */
 434         i_shift_offset   = DIM*shiftidx[iidx];
 435
 436         /* Load limits for loop over neighbors */
 437         j_index_start    = jindex[iidx];
 438         j_index_end      = jindex[iidx+1];
 439
 440         /* Get outer coordinate index */
 441         inr              = iinr[iidx];
 442         i_coord_offset   = DIM*inr;
 443
 444         /* Load i particle coords and add shift vector */
 445         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 446
 447         fix0             = _mm_setzero_ps();
 448         fiy0             = _mm_setzero_ps();
 449         fiz0             = _mm_setzero_ps();
 450
 451         /* Load parameters for i particles */
 452         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 453
 454         /* Start inner kernel loop */
 455         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 456         {
 457
 458             /* Get j neighbor index, and coordinate index */
 459             jnrA             = jjnr[jidx];
 460             jnrB             = jjnr[jidx+1];
 461             jnrC             = jjnr[jidx+2];
 462             jnrD             = jjnr[jidx+3];
 463             j_coord_offsetA  = DIM*jnrA;
 464             j_coord_offsetB  = DIM*jnrB;
 465             j_coord_offsetC  = DIM*jnrC;
 466             j_coord_offsetD  = DIM*jnrD;
 467
 468             /* load j atom coordinates */
 469             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                               x+j_coord_offsetC,x+j_coord_offsetD,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_ps(ix0,jx0);
 475             dy00             = _mm_sub_ps(iy0,jy0);
 476             dz00             = _mm_sub_ps(iz0,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 480
 481             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 482
 483             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 484
 485             /* Load parameters for j particles */
 486             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 487                                                               charge+jnrC+0,charge+jnrD+0);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             if (gmx_mm_any_lt(rsq00,rcutoff2))
 494             {
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq00             = _mm_mul_ps(iq0,jq0);
 498
 499             /* REACTION-FIELD ELECTROSTATICS */
 500             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 501
 502             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 503
 504             fscal            = felec;
 505
 506             fscal            = _mm_and_ps(fscal,cutoff_mask);
 507
 508              /* Update vectorial force */
 509             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 510             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 511             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 512
 513             fjptrA             = f+j_coord_offsetA;
 514             fjptrB             = f+j_coord_offsetB;
 515             fjptrC             = f+j_coord_offsetC;
 516             fjptrD             = f+j_coord_offsetD;
 517             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 518                                                    _mm_mul_ps(dx00,fscal),
 519                                                    _mm_mul_ps(dy00,fscal),
 520                                                    _mm_mul_ps(dz00,fscal));
 521
 522             }
 523
 524             /* Inner loop uses 33 flops */
 525         }
 526
 527         if(jidx<j_index_end)
 528         {
 529
 530             /* Get j neighbor index, and coordinate index */
 531             jnrlistA         = jjnr[jidx];
 532             jnrlistB         = jjnr[jidx+1];
 533             jnrlistC         = jjnr[jidx+2];
 534             jnrlistD         = jjnr[jidx+3];
 535             /* Sign of each element will be negative for non-real atoms.
 536              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 537              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 538              */
 539             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 540             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 541             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 542             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 543             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 544             j_coord_offsetA  = DIM*jnrA;
 545             j_coord_offsetB  = DIM*jnrB;
 546             j_coord_offsetC  = DIM*jnrC;
 547             j_coord_offsetD  = DIM*jnrD;
 548
 549             /* load j atom coordinates */
 550             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 551                                               x+j_coord_offsetC,x+j_coord_offsetD,
 552                                               &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm_sub_ps(ix0,jx0);
 556             dy00             = _mm_sub_ps(iy0,jy0);
 557             dz00             = _mm_sub_ps(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 563
 564             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 568                                                               charge+jnrC+0,charge+jnrD+0);
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             if (gmx_mm_any_lt(rsq00,rcutoff2))
 575             {
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq00             = _mm_mul_ps(iq0,jq0);
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 582
 583             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_and_ps(fscal,cutoff_mask);
 588
 589             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 590
 591              /* Update vectorial force */
 592             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 593             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 594             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 595
 596             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 597             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 598             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 599             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 600             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 601                                                    _mm_mul_ps(dx00,fscal),
 602                                                    _mm_mul_ps(dy00,fscal),
 603                                                    _mm_mul_ps(dz00,fscal));
 604
 605             }
 606
 607             /* Inner loop uses 33 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         /* Increment number of inner iterations */
 616         inneriter                  += j_index_end - j_index_start;
 617
 618         /* Outer loop uses 7 flops */
 619     }
 620
 621     /* Increment number of outer iterations */
 622     outeriter        += nri;
 623
 624     /* Update outer/inner flops */
 625
 626     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*33);
 627 }