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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_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     __m128i          ewitab;
  93     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113
 114     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 115     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 116     beta2            = _mm_mul_ps(beta,beta);
 117     beta3            = _mm_mul_ps(beta,beta2);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_ps();
 155         fiy0             = _mm_setzero_ps();
 156         fiz0             = _mm_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               x+j_coord_offsetC,x+j_coord_offsetD,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_ps(ix0,jx0);
 185             dy00             = _mm_sub_ps(iy0,jy0);
 186             dz00             = _mm_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 192
 193             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                               charge+jnrC+0,charge+jnrD+0);
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _mm_mul_ps(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_ps(iq0,jq0);
 207
 208             /* EWALD ELECTROSTATICS */
 209
 210             /* Analytical PME correction */
 211             zeta2            = _mm_mul_ps(beta2,rsq00);
 212             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 213             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 214             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 215             felec            = _mm_mul_ps(qq00,felec);
 216             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 217             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 218             velec            = _mm_mul_ps(qq00,velec);
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             velecsum         = _mm_add_ps(velecsum,velec);
 222
 223             fscal            = felec;
 224
 225              /* Update vectorial force */
 226             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 227             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 228             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 229
 230             fjptrA             = f+j_coord_offsetA;
 231             fjptrB             = f+j_coord_offsetB;
 232             fjptrC             = f+j_coord_offsetC;
 233             fjptrD             = f+j_coord_offsetD;
 234             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 235                                                    _mm_mul_ps(dx00,fscal),
 236                                                    _mm_mul_ps(dy00,fscal),
 237                                                    _mm_mul_ps(dz00,fscal));
 238
 239             /* Inner loop uses 29 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             /* Get j neighbor index, and coordinate index */
 246             jnrlistA         = jjnr[jidx];
 247             jnrlistB         = jjnr[jidx+1];
 248             jnrlistC         = jjnr[jidx+2];
 249             jnrlistD         = jjnr[jidx+3];
 250             /* Sign of each element will be negative for non-real atoms.
 251              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 252              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 253              */
 254             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 255             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 256             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 257             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 258             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                               x+j_coord_offsetC,x+j_coord_offsetD,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm_sub_ps(ix0,jx0);
 271             dy00             = _mm_sub_ps(iy0,jy0);
 272             dz00             = _mm_sub_ps(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 278
 279             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 283                                                               charge+jnrC+0,charge+jnrD+0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r00              = _mm_mul_ps(rsq00,rinv00);
 290             r00              = _mm_andnot_ps(dummy_mask,r00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm_mul_ps(iq0,jq0);
 294
 295             /* EWALD ELECTROSTATICS */
 296
 297             /* Analytical PME correction */
 298             zeta2            = _mm_mul_ps(beta2,rsq00);
 299             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 300             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 301             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 302             felec            = _mm_mul_ps(qq00,felec);
 303             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 304             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 305             velec            = _mm_mul_ps(qq00,velec);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_andnot_ps(dummy_mask,velec);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 314
 315              /* Update vectorial force */
 316             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 317             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 318             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 319
 320             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 321             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 322             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 323             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 324             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 325                                                    _mm_mul_ps(dx00,fscal),
 326                                                    _mm_mul_ps(dy00,fscal),
 327                                                    _mm_mul_ps(dz00,fscal));
 328
 329             /* Inner loop uses 30 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 335                                               f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 8 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*30);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_single
 356  * Electrostatics interaction: Ewald
 357  * VdW interaction:            None
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_single
 363                     (t_nblist                    * gmx_restrict       nlist,
 364                      rvec                        * gmx_restrict          xx,
 365                      rvec                        * gmx_restrict          ff,
 366                      t_forcerec                  * gmx_restrict          fr,
 367                      t_mdatoms                   * gmx_restrict     mdatoms,
 368                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 369                      t_nrnb                      * gmx_restrict        nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB,jnrC,jnrD;
 379     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 380     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 385     real             scratch[4*DIM];
 386     __m128           fscal,rcutoff,rcutoff2,jidxall;
 387     int              vdwioffset0;
 388     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 389     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 390     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 391     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 392     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 393     real             *charge;
 394     __m128i          ewitab;
 395     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 396     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 397     real             *ewtab;
 398     __m128           dummy_mask,cutoff_mask;
 399     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 400     __m128           one     = _mm_set1_ps(1.0);
 401     __m128           two     = _mm_set1_ps(2.0);
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     facel            = _mm_set1_ps(fr->epsfac);
 414     charge           = mdatoms->chargeA;
 415
 416     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 417     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 418     beta2            = _mm_mul_ps(beta,beta);
 419     beta3            = _mm_mul_ps(beta,beta2);
 420     ewtab            = fr->ic->tabq_coul_F;
 421     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 422     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 423
 424     /* Avoid stupid compiler warnings */
 425     jnrA = jnrB = jnrC = jnrD = 0;
 426     j_coord_offsetA = 0;
 427     j_coord_offsetB = 0;
 428     j_coord_offsetC = 0;
 429     j_coord_offsetD = 0;
 430
 431     outeriter        = 0;
 432     inneriter        = 0;
 433
 434     for(iidx=0;iidx<4*DIM;iidx++)
 435     {
 436         scratch[iidx] = 0.0;
 437     }
 438
 439     /* Start outer loop over neighborlists */
 440     for(iidx=0; iidx<nri; iidx++)
 441     {
 442         /* Load shift vector for this list */
 443         i_shift_offset   = DIM*shiftidx[iidx];
 444
 445         /* Load limits for loop over neighbors */
 446         j_index_start    = jindex[iidx];
 447         j_index_end      = jindex[iidx+1];
 448
 449         /* Get outer coordinate index */
 450         inr              = iinr[iidx];
 451         i_coord_offset   = DIM*inr;
 452
 453         /* Load i particle coords and add shift vector */
 454         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 455
 456         fix0             = _mm_setzero_ps();
 457         fiy0             = _mm_setzero_ps();
 458         fiz0             = _mm_setzero_ps();
 459
 460         /* Load parameters for i particles */
 461         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 462
 463         /* Start inner kernel loop */
 464         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrA             = jjnr[jidx];
 469             jnrB             = jjnr[jidx+1];
 470             jnrC             = jjnr[jidx+2];
 471             jnrD             = jjnr[jidx+3];
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474             j_coord_offsetC  = DIM*jnrC;
 475             j_coord_offsetD  = DIM*jnrD;
 476
 477             /* load j atom coordinates */
 478             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 479                                               x+j_coord_offsetC,x+j_coord_offsetD,
 480                                               &jx0,&jy0,&jz0);
 481
 482             /* Calculate displacement vector */
 483             dx00             = _mm_sub_ps(ix0,jx0);
 484             dy00             = _mm_sub_ps(iy0,jy0);
 485             dz00             = _mm_sub_ps(iz0,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 489
 490             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 491
 492             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r00              = _mm_mul_ps(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq00             = _mm_mul_ps(iq0,jq0);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Analytical PME correction */
 510             zeta2            = _mm_mul_ps(beta2,rsq00);
 511             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 512             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 513             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 514             felec            = _mm_mul_ps(qq00,felec);
 515
 516             fscal            = felec;
 517
 518              /* Update vectorial force */
 519             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 520             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 521             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 522
 523             fjptrA             = f+j_coord_offsetA;
 524             fjptrB             = f+j_coord_offsetB;
 525             fjptrC             = f+j_coord_offsetC;
 526             fjptrD             = f+j_coord_offsetD;
 527             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 528                                                    _mm_mul_ps(dx00,fscal),
 529                                                    _mm_mul_ps(dy00,fscal),
 530                                                    _mm_mul_ps(dz00,fscal));
 531
 532             /* Inner loop uses 28 flops */
 533         }
 534
 535         if(jidx<j_index_end)
 536         {
 537
 538             /* Get j neighbor index, and coordinate index */
 539             jnrlistA         = jjnr[jidx];
 540             jnrlistB         = jjnr[jidx+1];
 541             jnrlistC         = jjnr[jidx+2];
 542             jnrlistD         = jjnr[jidx+3];
 543             /* Sign of each element will be negative for non-real atoms.
 544              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 545              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 546              */
 547             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 548             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 549             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 550             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 551             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 552             j_coord_offsetA  = DIM*jnrA;
 553             j_coord_offsetB  = DIM*jnrB;
 554             j_coord_offsetC  = DIM*jnrC;
 555             j_coord_offsetD  = DIM*jnrD;
 556
 557             /* load j atom coordinates */
 558             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 559                                               x+j_coord_offsetC,x+j_coord_offsetD,
 560                                               &jx0,&jy0,&jz0);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm_sub_ps(ix0,jx0);
 564             dy00             = _mm_sub_ps(iy0,jy0);
 565             dz00             = _mm_sub_ps(iz0,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 569
 570             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 571
 572             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 573
 574             /* Load parameters for j particles */
 575             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 576                                                               charge+jnrC+0,charge+jnrD+0);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r00              = _mm_mul_ps(rsq00,rinv00);
 583             r00              = _mm_andnot_ps(dummy_mask,r00);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq00             = _mm_mul_ps(iq0,jq0);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Analytical PME correction */
 591             zeta2            = _mm_mul_ps(beta2,rsq00);
 592             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 593             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 594             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 595             felec            = _mm_mul_ps(qq00,felec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 600
 601              /* Update vectorial force */
 602             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 603             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 604             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 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,
 611                                                    _mm_mul_ps(dx00,fscal),
 612                                                    _mm_mul_ps(dy00,fscal),
 613                                                    _mm_mul_ps(dz00,fscal));
 614
 615             /* Inner loop uses 29 flops */
 616         }
 617
 618         /* End of innermost loop */
 619
 620         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 621                                               f+i_coord_offset,fshift+i_shift_offset);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 7 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*29);
 635 }