src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_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 "gmx_math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_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_ElecEwSh_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);
 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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_ps(ix0,jx0);
 190             dy00             = _mm_sub_ps(iy0,jy0);
 191             dz00             = _mm_sub_ps(iz0,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 195
 196             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 197
 198             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 199
 200             /* Load parameters for j particles */
 201             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 202                                                               charge+jnrC+0,charge+jnrD+0);
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             if (gmx_mm_any_lt(rsq00,rcutoff2))
 209             {
 210
 211             r00              = _mm_mul_ps(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_ps(iq0,jq0);
 215
 216             /* EWALD ELECTROSTATICS */
 217
 218             /* Analytical PME correction */
 219             zeta2            = _mm_mul_ps(beta2,rsq00);
 220             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 221             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 222             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 223             felec            = _mm_mul_ps(qq00,felec);
 224             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 225             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 226             velec            = _mm_mul_ps(qq00,velec);
 227
 228             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velec            = _mm_and_ps(velec,cutoff_mask);
 232             velecsum         = _mm_add_ps(velecsum,velec);
 233
 234             fscal            = felec;
 235
 236             fscal            = _mm_and_ps(fscal,cutoff_mask);
 237
 238              /* Update vectorial force */
 239             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 240             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 241             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 242
 243             fjptrA             = f+j_coord_offsetA;
 244             fjptrB             = f+j_coord_offsetB;
 245             fjptrC             = f+j_coord_offsetC;
 246             fjptrD             = f+j_coord_offsetD;
 247             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 248                                                    _mm_mul_ps(dx00,fscal),
 249                                                    _mm_mul_ps(dy00,fscal),
 250                                                    _mm_mul_ps(dz00,fscal));
 251
 252             }
 253
 254             /* Inner loop uses 33 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             /* Get j neighbor index, and coordinate index */
 261             jnrlistA         = jjnr[jidx];
 262             jnrlistB         = jjnr[jidx+1];
 263             jnrlistC         = jjnr[jidx+2];
 264             jnrlistD         = jjnr[jidx+3];
 265             /* Sign of each element will be negative for non-real atoms.
 266              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 267              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 268              */
 269             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 270             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 271             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 272             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 273             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                               x+j_coord_offsetC,x+j_coord_offsetD,
 282                                               &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm_sub_ps(ix0,jx0);
 286             dy00             = _mm_sub_ps(iy0,jy0);
 287             dz00             = _mm_sub_ps(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 291
 292             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 293
 294             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 295
 296             /* Load parameters for j particles */
 297             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 298                                                               charge+jnrC+0,charge+jnrD+0);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             if (gmx_mm_any_lt(rsq00,rcutoff2))
 305             {
 306
 307             r00              = _mm_mul_ps(rsq00,rinv00);
 308             r00              = _mm_andnot_ps(dummy_mask,r00);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm_mul_ps(iq0,jq0);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Analytical PME correction */
 316             zeta2            = _mm_mul_ps(beta2,rsq00);
 317             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 318             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 319             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 320             felec            = _mm_mul_ps(qq00,felec);
 321             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 322             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 323             velec            = _mm_mul_ps(qq00,velec);
 324
 325             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velec            = _mm_and_ps(velec,cutoff_mask);
 329             velec            = _mm_andnot_ps(dummy_mask,velec);
 330             velecsum         = _mm_add_ps(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             fscal            = _mm_and_ps(fscal,cutoff_mask);
 335
 336             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 337
 338              /* Update vectorial force */
 339             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 340             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 341             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 342
 343             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 344             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 345             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 346             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 347             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 348                                                    _mm_mul_ps(dx00,fscal),
 349                                                    _mm_mul_ps(dy00,fscal),
 350                                                    _mm_mul_ps(dz00,fscal));
 351
 352             }
 353
 354             /* Inner loop uses 34 flops */
 355         }
 356
 357         /* End of innermost loop */
 358
 359         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 360                                               f+i_coord_offset,fshift+i_shift_offset);
 361
 362         ggid                        = gid[iidx];
 363         /* Update potential energies */
 364         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 365
 366         /* Increment number of inner iterations */
 367         inneriter                  += j_index_end - j_index_start;
 368
 369         /* Outer loop uses 8 flops */
 370     }
 371
 372     /* Increment number of outer iterations */
 373     outeriter        += nri;
 374
 375     /* Update outer/inner flops */
 376
 377     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*34);
 378 }
 379 /*
 380  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 381  * Electrostatics interaction: Ewald
 382  * VdW interaction:            None
 383  * Geometry:                   Particle-Particle
 384  * Calculate force/pot:        Force
 385  */
 386 void
 387 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 388                     (t_nblist                    * gmx_restrict       nlist,
 389                      rvec                        * gmx_restrict          xx,
 390                      rvec                        * gmx_restrict          ff,
 391                      t_forcerec                  * gmx_restrict          fr,
 392                      t_mdatoms                   * gmx_restrict     mdatoms,
 393                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 394                      t_nrnb                      * gmx_restrict        nrnb)
 395 {
 396     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 397      * just 0 for non-waters.
 398      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 399      * jnr indices corresponding to data put in the four positions in the SIMD register.
 400      */
 401     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 402     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 403     int              jnrA,jnrB,jnrC,jnrD;
 404     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 405     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 406     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 407     real             rcutoff_scalar;
 408     real             *shiftvec,*fshift,*x,*f;
 409     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 410     real             scratch[4*DIM];
 411     __m128           fscal,rcutoff,rcutoff2,jidxall;
 412     int              vdwioffset0;
 413     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 414     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 415     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 416     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 417     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 418     real             *charge;
 419     __m128i          ewitab;
 420     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 421     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 422     real             *ewtab;
 423     __m128           dummy_mask,cutoff_mask;
 424     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 425     __m128           one     = _mm_set1_ps(1.0);
 426     __m128           two     = _mm_set1_ps(2.0);
 427     x                = xx[0];
 428     f                = ff[0];
 429
 430     nri              = nlist->nri;
 431     iinr             = nlist->iinr;
 432     jindex           = nlist->jindex;
 433     jjnr             = nlist->jjnr;
 434     shiftidx         = nlist->shift;
 435     gid              = nlist->gid;
 436     shiftvec         = fr->shift_vec[0];
 437     fshift           = fr->fshift[0];
 438     facel            = _mm_set1_ps(fr->epsfac);
 439     charge           = mdatoms->chargeA;
 440
 441     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 442     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 443     beta2            = _mm_mul_ps(beta,beta);
 444     beta3            = _mm_mul_ps(beta,beta2);
 445     ewtab            = fr->ic->tabq_coul_F;
 446     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 447     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 448
 449     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 450     rcutoff_scalar   = fr->rcoulomb;
 451     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 452     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 453
 454     /* Avoid stupid compiler warnings */
 455     jnrA = jnrB = jnrC = jnrD = 0;
 456     j_coord_offsetA = 0;
 457     j_coord_offsetB = 0;
 458     j_coord_offsetC = 0;
 459     j_coord_offsetD = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     for(iidx=0;iidx<4*DIM;iidx++)
 465     {
 466         scratch[iidx] = 0.0;
 467     }
 468
 469     /* Start outer loop over neighborlists */
 470     for(iidx=0; iidx<nri; iidx++)
 471     {
 472         /* Load shift vector for this list */
 473         i_shift_offset   = DIM*shiftidx[iidx];
 474
 475         /* Load limits for loop over neighbors */
 476         j_index_start    = jindex[iidx];
 477         j_index_end      = jindex[iidx+1];
 478
 479         /* Get outer coordinate index */
 480         inr              = iinr[iidx];
 481         i_coord_offset   = DIM*inr;
 482
 483         /* Load i particle coords and add shift vector */
 484         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 485
 486         fix0             = _mm_setzero_ps();
 487         fiy0             = _mm_setzero_ps();
 488         fiz0             = _mm_setzero_ps();
 489
 490         /* Load parameters for i particles */
 491         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             jnrC             = jjnr[jidx+2];
 501             jnrD             = jjnr[jidx+3];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504             j_coord_offsetC  = DIM*jnrC;
 505             j_coord_offsetD  = DIM*jnrD;
 506
 507             /* load j atom coordinates */
 508             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                               x+j_coord_offsetC,x+j_coord_offsetD,
 510                                               &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _mm_sub_ps(ix0,jx0);
 514             dy00             = _mm_sub_ps(iy0,jy0);
 515             dz00             = _mm_sub_ps(iz0,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 519
 520             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 521
 522             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 523
 524             /* Load parameters for j particles */
 525             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 526                                                               charge+jnrC+0,charge+jnrD+0);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_mm_any_lt(rsq00,rcutoff2))
 533             {
 534
 535             r00              = _mm_mul_ps(rsq00,rinv00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _mm_mul_ps(iq0,jq0);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Analytical PME correction */
 543             zeta2            = _mm_mul_ps(beta2,rsq00);
 544             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 545             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 546             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 547             felec            = _mm_mul_ps(qq00,felec);
 548
 549             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm_and_ps(fscal,cutoff_mask);
 554
 555              /* Update vectorial force */
 556             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 557             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 558             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 559
 560             fjptrA             = f+j_coord_offsetA;
 561             fjptrB             = f+j_coord_offsetB;
 562             fjptrC             = f+j_coord_offsetC;
 563             fjptrD             = f+j_coord_offsetD;
 564             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 565                                                    _mm_mul_ps(dx00,fscal),
 566                                                    _mm_mul_ps(dy00,fscal),
 567                                                    _mm_mul_ps(dz00,fscal));
 568
 569             }
 570
 571             /* Inner loop uses 31 flops */
 572         }
 573
 574         if(jidx<j_index_end)
 575         {
 576
 577             /* Get j neighbor index, and coordinate index */
 578             jnrlistA         = jjnr[jidx];
 579             jnrlistB         = jjnr[jidx+1];
 580             jnrlistC         = jjnr[jidx+2];
 581             jnrlistD         = jjnr[jidx+3];
 582             /* Sign of each element will be negative for non-real atoms.
 583              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 584              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 585              */
 586             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 587             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 588             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 589             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 590             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 591             j_coord_offsetA  = DIM*jnrA;
 592             j_coord_offsetB  = DIM*jnrB;
 593             j_coord_offsetC  = DIM*jnrC;
 594             j_coord_offsetD  = DIM*jnrD;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 598                                               x+j_coord_offsetC,x+j_coord_offsetD,
 599                                               &jx0,&jy0,&jz0);
 600
 601             /* Calculate displacement vector */
 602             dx00             = _mm_sub_ps(ix0,jx0);
 603             dy00             = _mm_sub_ps(iy0,jy0);
 604             dz00             = _mm_sub_ps(iz0,jz0);
 605
 606             /* Calculate squared distance and things based on it */
 607             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 608
 609             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 610
 611             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 612
 613             /* Load parameters for j particles */
 614             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 615                                                               charge+jnrC+0,charge+jnrD+0);
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm_any_lt(rsq00,rcutoff2))
 622             {
 623
 624             r00              = _mm_mul_ps(rsq00,rinv00);
 625             r00              = _mm_andnot_ps(dummy_mask,r00);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq00             = _mm_mul_ps(iq0,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Analytical PME correction */
 633             zeta2            = _mm_mul_ps(beta2,rsq00);
 634             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 635             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 636             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 637             felec            = _mm_mul_ps(qq00,felec);
 638
 639             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm_and_ps(fscal,cutoff_mask);
 644
 645             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 646
 647              /* Update vectorial force */
 648             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 649             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 650             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 651
 652             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 653             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 654             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 655             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 656             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 657                                                    _mm_mul_ps(dx00,fscal),
 658                                                    _mm_mul_ps(dy00,fscal),
 659                                                    _mm_mul_ps(dz00,fscal));
 660
 661             }
 662
 663             /* Inner loop uses 32 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*32);
 683 }