src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015, 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 sse2_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/math/vec.h"
  46 #include "gromacs/legacyheaders/nrnb.h"
  47
  48 #include "gromacs/simd/math_x86_sse2_single.h"
  49 #include "kernelutil_x86_sse2_single.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_sse2_single
  53  * Electrostatics interaction: ReactionField
  54  * VdW interaction:            None
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_sse2_single
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB,jnrC,jnrD;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     int              vdwioffset0;
  85     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     krf              = _mm_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm_set1_ps(fr->ic->c_rf);
 111
 112     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 113     rcutoff_scalar   = fr->rcoulomb;
 114     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 115     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 187
 188             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 192                                                               charge+jnrC+0,charge+jnrD+0);
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             if (gmx_mm_any_lt(rsq00,rcutoff2))
 199             {
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             qq00             = _mm_mul_ps(iq0,jq0);
 203
 204             /* REACTION-FIELD ELECTROSTATICS */
 205             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 206             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 207
 208             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             velec            = _mm_and_ps(velec,cutoff_mask);
 212             velecsum         = _mm_add_ps(velecsum,velec);
 213
 214             fscal            = felec;
 215
 216             fscal            = _mm_and_ps(fscal,cutoff_mask);
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_ps(fscal,dx00);
 220             ty               = _mm_mul_ps(fscal,dy00);
 221             tz               = _mm_mul_ps(fscal,dz00);
 222
 223             /* Update vectorial force */
 224             fix0             = _mm_add_ps(fix0,tx);
 225             fiy0             = _mm_add_ps(fiy0,ty);
 226             fiz0             = _mm_add_ps(fiz0,tz);
 227
 228             fjptrA             = f+j_coord_offsetA;
 229             fjptrB             = f+j_coord_offsetB;
 230             fjptrC             = f+j_coord_offsetC;
 231             fjptrD             = f+j_coord_offsetD;
 232             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 233
 234             }
 235
 236             /* Inner loop uses 36 flops */
 237         }
 238
 239         if(jidx<j_index_end)
 240         {
 241
 242             /* Get j neighbor index, and coordinate index */
 243             jnrlistA         = jjnr[jidx];
 244             jnrlistB         = jjnr[jidx+1];
 245             jnrlistC         = jjnr[jidx+2];
 246             jnrlistD         = jjnr[jidx+3];
 247             /* Sign of each element will be negative for non-real atoms.
 248              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 249              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 250              */
 251             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 252             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 253             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 254             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 255             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 256             j_coord_offsetA  = DIM*jnrA;
 257             j_coord_offsetB  = DIM*jnrB;
 258             j_coord_offsetC  = DIM*jnrC;
 259             j_coord_offsetD  = DIM*jnrD;
 260
 261             /* load j atom coordinates */
 262             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 263                                               x+j_coord_offsetC,x+j_coord_offsetD,
 264                                               &jx0,&jy0,&jz0);
 265
 266             /* Calculate displacement vector */
 267             dx00             = _mm_sub_ps(ix0,jx0);
 268             dy00             = _mm_sub_ps(iy0,jy0);
 269             dz00             = _mm_sub_ps(iz0,jz0);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 273
 274             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 275
 276             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 277
 278             /* Load parameters for j particles */
 279             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 280                                                               charge+jnrC+0,charge+jnrD+0);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             if (gmx_mm_any_lt(rsq00,rcutoff2))
 287             {
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq00             = _mm_mul_ps(iq0,jq0);
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 294             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 295
 296             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_ps(velec,cutoff_mask);
 300             velec            = _mm_andnot_ps(dummy_mask,velec);
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             fscal            = _mm_and_ps(fscal,cutoff_mask);
 306
 307             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_ps(fscal,dx00);
 311             ty               = _mm_mul_ps(fscal,dy00);
 312             tz               = _mm_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_add_ps(fix0,tx);
 316             fiy0             = _mm_add_ps(fiy0,ty);
 317             fiz0             = _mm_add_ps(fiz0,tz);
 318
 319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 323             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 324
 325             }
 326
 327             /* Inner loop uses 36 flops */
 328         }
 329
 330         /* End of innermost loop */
 331
 332         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 333                                               f+i_coord_offset,fshift+i_shift_offset);
 334
 335         ggid                        = gid[iidx];
 336         /* Update potential energies */
 337         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 338
 339         /* Increment number of inner iterations */
 340         inneriter                  += j_index_end - j_index_start;
 341
 342         /* Outer loop uses 8 flops */
 343     }
 344
 345     /* Increment number of outer iterations */
 346     outeriter        += nri;
 347
 348     /* Update outer/inner flops */
 349
 350     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse2_single
 354  * Electrostatics interaction: ReactionField
 355  * VdW interaction:            None
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse2_single
 361                     (t_nblist                    * gmx_restrict       nlist,
 362                      rvec                        * gmx_restrict          xx,
 363                      rvec                        * gmx_restrict          ff,
 364                      t_forcerec                  * gmx_restrict          fr,
 365                      t_mdatoms                   * gmx_restrict     mdatoms,
 366                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 367                      t_nrnb                      * gmx_restrict        nrnb)
 368 {
 369     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 370      * just 0 for non-waters.
 371      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 372      * jnr indices corresponding to data put in the four positions in the SIMD register.
 373      */
 374     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 375     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 376     int              jnrA,jnrB,jnrC,jnrD;
 377     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 378     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 379     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 380     real             rcutoff_scalar;
 381     real             *shiftvec,*fshift,*x,*f;
 382     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 383     real             scratch[4*DIM];
 384     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     int              vdwioffset0;
 386     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 388     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 391     real             *charge;
 392     __m128           dummy_mask,cutoff_mask;
 393     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 394     __m128           one     = _mm_set1_ps(1.0);
 395     __m128           two     = _mm_set1_ps(2.0);
 396     x                = xx[0];
 397     f                = ff[0];
 398
 399     nri              = nlist->nri;
 400     iinr             = nlist->iinr;
 401     jindex           = nlist->jindex;
 402     jjnr             = nlist->jjnr;
 403     shiftidx         = nlist->shift;
 404     gid              = nlist->gid;
 405     shiftvec         = fr->shift_vec[0];
 406     fshift           = fr->fshift[0];
 407     facel            = _mm_set1_ps(fr->epsfac);
 408     charge           = mdatoms->chargeA;
 409     krf              = _mm_set1_ps(fr->ic->k_rf);
 410     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 411     crf              = _mm_set1_ps(fr->ic->c_rf);
 412
 413     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 414     rcutoff_scalar   = fr->rcoulomb;
 415     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 416     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 417
 418     /* Avoid stupid compiler warnings */
 419     jnrA = jnrB = jnrC = jnrD = 0;
 420     j_coord_offsetA = 0;
 421     j_coord_offsetB = 0;
 422     j_coord_offsetC = 0;
 423     j_coord_offsetD = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     for(iidx=0;iidx<4*DIM;iidx++)
 429     {
 430         scratch[iidx] = 0.0;
 431     }
 432
 433     /* Start outer loop over neighborlists */
 434     for(iidx=0; iidx<nri; iidx++)
 435     {
 436         /* Load shift vector for this list */
 437         i_shift_offset   = DIM*shiftidx[iidx];
 438
 439         /* Load limits for loop over neighbors */
 440         j_index_start    = jindex[iidx];
 441         j_index_end      = jindex[iidx+1];
 442
 443         /* Get outer coordinate index */
 444         inr              = iinr[iidx];
 445         i_coord_offset   = DIM*inr;
 446
 447         /* Load i particle coords and add shift vector */
 448         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 449
 450         fix0             = _mm_setzero_ps();
 451         fiy0             = _mm_setzero_ps();
 452         fiz0             = _mm_setzero_ps();
 453
 454         /* Load parameters for i particles */
 455         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 456
 457         /* Start inner kernel loop */
 458         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrA             = jjnr[jidx];
 463             jnrB             = jjnr[jidx+1];
 464             jnrC             = jjnr[jidx+2];
 465             jnrD             = jjnr[jidx+3];
 466             j_coord_offsetA  = DIM*jnrA;
 467             j_coord_offsetB  = DIM*jnrB;
 468             j_coord_offsetC  = DIM*jnrC;
 469             j_coord_offsetD  = DIM*jnrD;
 470
 471             /* load j atom coordinates */
 472             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 473                                               x+j_coord_offsetC,x+j_coord_offsetD,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _mm_sub_ps(ix0,jx0);
 478             dy00             = _mm_sub_ps(iy0,jy0);
 479             dz00             = _mm_sub_ps(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 483
 484             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 485
 486             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 490                                                               charge+jnrC+0,charge+jnrD+0);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             if (gmx_mm_any_lt(rsq00,rcutoff2))
 497             {
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq00             = _mm_mul_ps(iq0,jq0);
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 504
 505             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_and_ps(fscal,cutoff_mask);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx00);
 513             ty               = _mm_mul_ps(fscal,dy00);
 514             tz               = _mm_mul_ps(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_ps(fix0,tx);
 518             fiy0             = _mm_add_ps(fiy0,ty);
 519             fiz0             = _mm_add_ps(fiz0,tz);
 520
 521             fjptrA             = f+j_coord_offsetA;
 522             fjptrB             = f+j_coord_offsetB;
 523             fjptrC             = f+j_coord_offsetC;
 524             fjptrD             = f+j_coord_offsetD;
 525             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 526
 527             }
 528
 529             /* Inner loop uses 30 flops */
 530         }
 531
 532         if(jidx<j_index_end)
 533         {
 534
 535             /* Get j neighbor index, and coordinate index */
 536             jnrlistA         = jjnr[jidx];
 537             jnrlistB         = jjnr[jidx+1];
 538             jnrlistC         = jjnr[jidx+2];
 539             jnrlistD         = jjnr[jidx+3];
 540             /* Sign of each element will be negative for non-real atoms.
 541              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 542              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 543              */
 544             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 545             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 546             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 547             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 548             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 549             j_coord_offsetA  = DIM*jnrA;
 550             j_coord_offsetB  = DIM*jnrB;
 551             j_coord_offsetC  = DIM*jnrC;
 552             j_coord_offsetD  = DIM*jnrD;
 553
 554             /* load j atom coordinates */
 555             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 556                                               x+j_coord_offsetC,x+j_coord_offsetD,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm_sub_ps(ix0,jx0);
 561             dy00             = _mm_sub_ps(iy0,jy0);
 562             dz00             = _mm_sub_ps(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 566
 567             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 568
 569             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 570
 571             /* Load parameters for j particles */
 572             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 573                                                               charge+jnrC+0,charge+jnrD+0);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq00,rcutoff2))
 580             {
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq00             = _mm_mul_ps(iq0,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 587
 588             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_and_ps(fscal,cutoff_mask);
 593
 594             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx00);
 598             ty               = _mm_mul_ps(fscal,dy00);
 599             tz               = _mm_mul_ps(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm_add_ps(fix0,tx);
 603             fiy0             = _mm_add_ps(fiy0,ty);
 604             fiz0             = _mm_add_ps(fiz0,tz);
 605
 606             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 607             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 608             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 609             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 610             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 611
 612             }
 613
 614             /* Inner loop uses 30 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 620                                               f+i_coord_offset,fshift+i_shift_offset);
 621
 622         /* Increment number of inner iterations */
 623         inneriter                  += j_index_end - j_index_start;
 624
 625         /* Outer loop uses 7 flops */
 626     }
 627
 628     /* Increment number of outer iterations */
 629     outeriter        += nri;
 630
 631     /* Update outer/inner flops */
 632
 633     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 634 }