src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_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     __m128i          ewitab;
  91     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  92     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  93     real             *ewtab;
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 113     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 114     beta2            = _mm_mul_ps(beta,beta);
 115     beta3            = _mm_mul_ps(beta,beta2);
 116     ewtab            = fr->ic->tabq_coul_FDV0;
 117     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 118     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               x+j_coord_offsetC,x+j_coord_offsetD,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_ps(ix0,jx0);
 183             dy00             = _mm_sub_ps(iy0,jy0);
 184             dz00             = _mm_sub_ps(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 190
 191             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 195                                                               charge+jnrC+0,charge+jnrD+0);
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             r00              = _mm_mul_ps(rsq00,rinv00);
 202
 203             /* Compute parameters for interactions between i and j atoms */
 204             qq00             = _mm_mul_ps(iq0,jq0);
 205
 206             /* EWALD ELECTROSTATICS */
 207
 208             /* Analytical PME correction */
 209             zeta2            = _mm_mul_ps(beta2,rsq00);
 210             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 211             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 212             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 213             felec            = _mm_mul_ps(qq00,felec);
 214             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 215             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 216             velec            = _mm_mul_ps(qq00,velec);
 217
 218             /* Update potential sum for this i atom from the interaction with this j atom. */
 219             velecsum         = _mm_add_ps(velecsum,velec);
 220
 221             fscal            = felec;
 222
 223              /* Update vectorial force */
 224             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 225             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 226             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 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,
 233                                                    _mm_mul_ps(dx00,fscal),
 234                                                    _mm_mul_ps(dy00,fscal),
 235                                                    _mm_mul_ps(dz00,fscal));
 236
 237             /* Inner loop uses 29 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             r00              = _mm_mul_ps(rsq00,rinv00);
 288             r00              = _mm_andnot_ps(dummy_mask,r00);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq00             = _mm_mul_ps(iq0,jq0);
 292
 293             /* EWALD ELECTROSTATICS */
 294
 295             /* Analytical PME correction */
 296             zeta2            = _mm_mul_ps(beta2,rsq00);
 297             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 298             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 299             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 300             felec            = _mm_mul_ps(qq00,felec);
 301             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 302             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 303             velec            = _mm_mul_ps(qq00,velec);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _mm_andnot_ps(dummy_mask,velec);
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 312
 313              /* Update vectorial force */
 314             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 315             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 316             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 317
 318             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 319             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 320             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 321             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 322             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 323                                                    _mm_mul_ps(dx00,fscal),
 324                                                    _mm_mul_ps(dy00,fscal),
 325                                                    _mm_mul_ps(dz00,fscal));
 326
 327             /* Inner loop uses 30 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*30);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_single
 354  * Electrostatics interaction: Ewald
 355  * VdW interaction:            None
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_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 AVX_128, 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           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     __m128i          ewitab;
 393     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 394     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 395     real             *ewtab;
 396     __m128           dummy_mask,cutoff_mask;
 397     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 398     __m128           one     = _mm_set1_ps(1.0);
 399     __m128           two     = _mm_set1_ps(2.0);
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     facel            = _mm_set1_ps(fr->epsfac);
 412     charge           = mdatoms->chargeA;
 413
 414     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 415     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 416     beta2            = _mm_mul_ps(beta,beta);
 417     beta3            = _mm_mul_ps(beta,beta2);
 418     ewtab            = fr->ic->tabq_coul_F;
 419     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 420     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 421
 422     /* Avoid stupid compiler warnings */
 423     jnrA = jnrB = jnrC = jnrD = 0;
 424     j_coord_offsetA = 0;
 425     j_coord_offsetB = 0;
 426     j_coord_offsetC = 0;
 427     j_coord_offsetD = 0;
 428
 429     outeriter        = 0;
 430     inneriter        = 0;
 431
 432     for(iidx=0;iidx<4*DIM;iidx++)
 433     {
 434         scratch[iidx] = 0.0;
 435     }
 436
 437     /* Start outer loop over neighborlists */
 438     for(iidx=0; iidx<nri; iidx++)
 439     {
 440         /* Load shift vector for this list */
 441         i_shift_offset   = DIM*shiftidx[iidx];
 442
 443         /* Load limits for loop over neighbors */
 444         j_index_start    = jindex[iidx];
 445         j_index_end      = jindex[iidx+1];
 446
 447         /* Get outer coordinate index */
 448         inr              = iinr[iidx];
 449         i_coord_offset   = DIM*inr;
 450
 451         /* Load i particle coords and add shift vector */
 452         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 453
 454         fix0             = _mm_setzero_ps();
 455         fiy0             = _mm_setzero_ps();
 456         fiz0             = _mm_setzero_ps();
 457
 458         /* Load parameters for i particles */
 459         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 460
 461         /* Start inner kernel loop */
 462         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrA             = jjnr[jidx];
 467             jnrB             = jjnr[jidx+1];
 468             jnrC             = jjnr[jidx+2];
 469             jnrD             = jjnr[jidx+3];
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472             j_coord_offsetC  = DIM*jnrC;
 473             j_coord_offsetD  = DIM*jnrD;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               x+j_coord_offsetC,x+j_coord_offsetD,
 478                                               &jx0,&jy0,&jz0);
 479
 480             /* Calculate displacement vector */
 481             dx00             = _mm_sub_ps(ix0,jx0);
 482             dy00             = _mm_sub_ps(iy0,jy0);
 483             dz00             = _mm_sub_ps(iz0,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 487
 488             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 489
 490             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 491
 492             /* Load parameters for j particles */
 493             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 494                                                               charge+jnrC+0,charge+jnrD+0);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_ps(rsq00,rinv00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm_mul_ps(iq0,jq0);
 504
 505             /* EWALD ELECTROSTATICS */
 506
 507             /* Analytical PME correction */
 508             zeta2            = _mm_mul_ps(beta2,rsq00);
 509             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 510             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 511             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 512             felec            = _mm_mul_ps(qq00,felec);
 513
 514             fscal            = felec;
 515
 516              /* Update vectorial force */
 517             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 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,
 526                                                    _mm_mul_ps(dx00,fscal),
 527                                                    _mm_mul_ps(dy00,fscal),
 528                                                    _mm_mul_ps(dz00,fscal));
 529
 530             /* Inner loop uses 28 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             r00              = _mm_mul_ps(rsq00,rinv00);
 581             r00              = _mm_andnot_ps(dummy_mask,r00);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq00             = _mm_mul_ps(iq0,jq0);
 585
 586             /* EWALD ELECTROSTATICS */
 587
 588             /* Analytical PME correction */
 589             zeta2            = _mm_mul_ps(beta2,rsq00);
 590             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 591             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 592             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 593             felec            = _mm_mul_ps(qq00,felec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 598
 599              /* Update vectorial force */
 600             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 601             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 602             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 609                                                    _mm_mul_ps(dx00,fscal),
 610                                                    _mm_mul_ps(dy00,fscal),
 611                                                    _mm_mul_ps(dz00,fscal));
 612
 613             /* Inner loop uses 29 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 619                                               f+i_coord_offset,fshift+i_shift_offset);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 7 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*29);
 633 }