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) 2012,2013,2014, 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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecEwSh_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_ElecEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 121     rcutoff_scalar   = fr->rcoulomb;
 122     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 123     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm_setzero_ps();
 158         fiy0             = _mm_setzero_ps();
 159         fiz0             = _mm_setzero_ps();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               x+j_coord_offsetC,x+j_coord_offsetD,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_ps(ix0,jx0);
 188             dy00             = _mm_sub_ps(iy0,jy0);
 189             dz00             = _mm_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 195
 196             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             if (gmx_mm_any_lt(rsq00,rcutoff2))
 207             {
 208
 209             r00              = _mm_mul_ps(rsq00,rinv00);
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm_mul_ps(iq0,jq0);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Analytical PME correction */
 217             zeta2            = _mm_mul_ps(beta2,rsq00);
 218             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 219             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 220             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 221             felec            = _mm_mul_ps(qq00,felec);
 222             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 223             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 224             velec            = _mm_mul_ps(qq00,velec);
 225
 226             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velec            = _mm_and_ps(velec,cutoff_mask);
 230             velecsum         = _mm_add_ps(velecsum,velec);
 231
 232             fscal            = felec;
 233
 234             fscal            = _mm_and_ps(fscal,cutoff_mask);
 235
 236              /* Update vectorial force */
 237             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 238             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 239             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 240
 241             fjptrA             = f+j_coord_offsetA;
 242             fjptrB             = f+j_coord_offsetB;
 243             fjptrC             = f+j_coord_offsetC;
 244             fjptrD             = f+j_coord_offsetD;
 245             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 246                                                    _mm_mul_ps(dx00,fscal),
 247                                                    _mm_mul_ps(dy00,fscal),
 248                                                    _mm_mul_ps(dz00,fscal));
 249
 250             }
 251
 252             /* Inner loop uses 33 flops */
 253         }
 254
 255         if(jidx<j_index_end)
 256         {
 257
 258             /* Get j neighbor index, and coordinate index */
 259             jnrlistA         = jjnr[jidx];
 260             jnrlistB         = jjnr[jidx+1];
 261             jnrlistC         = jjnr[jidx+2];
 262             jnrlistD         = jjnr[jidx+3];
 263             /* Sign of each element will be negative for non-real atoms.
 264              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 265              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 266              */
 267             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276
 277             /* load j atom coordinates */
 278             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 279                                               x+j_coord_offsetC,x+j_coord_offsetD,
 280                                               &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm_sub_ps(ix0,jx0);
 284             dy00             = _mm_sub_ps(iy0,jy0);
 285             dz00             = _mm_sub_ps(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 289
 290             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 291
 292             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 293
 294             /* Load parameters for j particles */
 295             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 296                                                               charge+jnrC+0,charge+jnrD+0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             if (gmx_mm_any_lt(rsq00,rcutoff2))
 303             {
 304
 305             r00              = _mm_mul_ps(rsq00,rinv00);
 306             r00              = _mm_andnot_ps(dummy_mask,r00);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq00             = _mm_mul_ps(iq0,jq0);
 310
 311             /* EWALD ELECTROSTATICS */
 312
 313             /* Analytical PME correction */
 314             zeta2            = _mm_mul_ps(beta2,rsq00);
 315             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 316             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 317             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 318             felec            = _mm_mul_ps(qq00,felec);
 319             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 320             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 321             velec            = _mm_mul_ps(qq00,velec);
 322
 323             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_and_ps(velec,cutoff_mask);
 327             velec            = _mm_andnot_ps(dummy_mask,velec);
 328             velecsum         = _mm_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             fscal            = _mm_and_ps(fscal,cutoff_mask);
 333
 334             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 335
 336              /* Update vectorial force */
 337             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 338             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 339             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 340
 341             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 342             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 343             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 344             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 346                                                    _mm_mul_ps(dx00,fscal),
 347                                                    _mm_mul_ps(dy00,fscal),
 348                                                    _mm_mul_ps(dz00,fscal));
 349
 350             }
 351
 352             /* Inner loop uses 34 flops */
 353         }
 354
 355         /* End of innermost loop */
 356
 357         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 358                                               f+i_coord_offset,fshift+i_shift_offset);
 359
 360         ggid                        = gid[iidx];
 361         /* Update potential energies */
 362         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 363
 364         /* Increment number of inner iterations */
 365         inneriter                  += j_index_end - j_index_start;
 366
 367         /* Outer loop uses 8 flops */
 368     }
 369
 370     /* Increment number of outer iterations */
 371     outeriter        += nri;
 372
 373     /* Update outer/inner flops */
 374
 375     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*34);
 376 }
 377 /*
 378  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 379  * Electrostatics interaction: Ewald
 380  * VdW interaction:            None
 381  * Geometry:                   Particle-Particle
 382  * Calculate force/pot:        Force
 383  */
 384 void
 385 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 386                     (t_nblist                    * gmx_restrict       nlist,
 387                      rvec                        * gmx_restrict          xx,
 388                      rvec                        * gmx_restrict          ff,
 389                      t_forcerec                  * gmx_restrict          fr,
 390                      t_mdatoms                   * gmx_restrict     mdatoms,
 391                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 392                      t_nrnb                      * gmx_restrict        nrnb)
 393 {
 394     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 395      * just 0 for non-waters.
 396      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 397      * jnr indices corresponding to data put in the four positions in the SIMD register.
 398      */
 399     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 400     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 401     int              jnrA,jnrB,jnrC,jnrD;
 402     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 403     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 404     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 405     real             rcutoff_scalar;
 406     real             *shiftvec,*fshift,*x,*f;
 407     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 408     real             scratch[4*DIM];
 409     __m128           fscal,rcutoff,rcutoff2,jidxall;
 410     int              vdwioffset0;
 411     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 412     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 413     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 414     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 415     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 416     real             *charge;
 417     __m128i          ewitab;
 418     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 419     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 420     real             *ewtab;
 421     __m128           dummy_mask,cutoff_mask;
 422     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 423     __m128           one     = _mm_set1_ps(1.0);
 424     __m128           two     = _mm_set1_ps(2.0);
 425     x                = xx[0];
 426     f                = ff[0];
 427
 428     nri              = nlist->nri;
 429     iinr             = nlist->iinr;
 430     jindex           = nlist->jindex;
 431     jjnr             = nlist->jjnr;
 432     shiftidx         = nlist->shift;
 433     gid              = nlist->gid;
 434     shiftvec         = fr->shift_vec[0];
 435     fshift           = fr->fshift[0];
 436     facel            = _mm_set1_ps(fr->epsfac);
 437     charge           = mdatoms->chargeA;
 438
 439     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 440     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 441     beta2            = _mm_mul_ps(beta,beta);
 442     beta3            = _mm_mul_ps(beta,beta2);
 443     ewtab            = fr->ic->tabq_coul_F;
 444     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 445     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 446
 447     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 448     rcutoff_scalar   = fr->rcoulomb;
 449     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 450     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 451
 452     /* Avoid stupid compiler warnings */
 453     jnrA = jnrB = jnrC = jnrD = 0;
 454     j_coord_offsetA = 0;
 455     j_coord_offsetB = 0;
 456     j_coord_offsetC = 0;
 457     j_coord_offsetD = 0;
 458
 459     outeriter        = 0;
 460     inneriter        = 0;
 461
 462     for(iidx=0;iidx<4*DIM;iidx++)
 463     {
 464         scratch[iidx] = 0.0;
 465     }
 466
 467     /* Start outer loop over neighborlists */
 468     for(iidx=0; iidx<nri; iidx++)
 469     {
 470         /* Load shift vector for this list */
 471         i_shift_offset   = DIM*shiftidx[iidx];
 472
 473         /* Load limits for loop over neighbors */
 474         j_index_start    = jindex[iidx];
 475         j_index_end      = jindex[iidx+1];
 476
 477         /* Get outer coordinate index */
 478         inr              = iinr[iidx];
 479         i_coord_offset   = DIM*inr;
 480
 481         /* Load i particle coords and add shift vector */
 482         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 483
 484         fix0             = _mm_setzero_ps();
 485         fiy0             = _mm_setzero_ps();
 486         fiz0             = _mm_setzero_ps();
 487
 488         /* Load parameters for i particles */
 489         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 490
 491         /* Start inner kernel loop */
 492         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrA             = jjnr[jidx];
 497             jnrB             = jjnr[jidx+1];
 498             jnrC             = jjnr[jidx+2];
 499             jnrD             = jjnr[jidx+3];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502             j_coord_offsetC  = DIM*jnrC;
 503             j_coord_offsetD  = DIM*jnrD;
 504
 505             /* load j atom coordinates */
 506             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               x+j_coord_offsetC,x+j_coord_offsetD,
 508                                               &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm_sub_ps(ix0,jx0);
 512             dy00             = _mm_sub_ps(iy0,jy0);
 513             dz00             = _mm_sub_ps(iz0,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 517
 518             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 519
 520             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                               charge+jnrC+0,charge+jnrD+0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             if (gmx_mm_any_lt(rsq00,rcutoff2))
 531             {
 532
 533             r00              = _mm_mul_ps(rsq00,rinv00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq00             = _mm_mul_ps(iq0,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Analytical PME correction */
 541             zeta2            = _mm_mul_ps(beta2,rsq00);
 542             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 543             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 544             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 545             felec            = _mm_mul_ps(qq00,felec);
 546
 547             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm_and_ps(fscal,cutoff_mask);
 552
 553              /* Update vectorial force */
 554             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 555             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 556             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 557
 558             fjptrA             = f+j_coord_offsetA;
 559             fjptrB             = f+j_coord_offsetB;
 560             fjptrC             = f+j_coord_offsetC;
 561             fjptrD             = f+j_coord_offsetD;
 562             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 563                                                    _mm_mul_ps(dx00,fscal),
 564                                                    _mm_mul_ps(dy00,fscal),
 565                                                    _mm_mul_ps(dz00,fscal));
 566
 567             }
 568
 569             /* Inner loop uses 31 flops */
 570         }
 571
 572         if(jidx<j_index_end)
 573         {
 574
 575             /* Get j neighbor index, and coordinate index */
 576             jnrlistA         = jjnr[jidx];
 577             jnrlistB         = jjnr[jidx+1];
 578             jnrlistC         = jjnr[jidx+2];
 579             jnrlistD         = jjnr[jidx+3];
 580             /* Sign of each element will be negative for non-real atoms.
 581              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 582              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 583              */
 584             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 585             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 586             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 587             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 588             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 589             j_coord_offsetA  = DIM*jnrA;
 590             j_coord_offsetB  = DIM*jnrB;
 591             j_coord_offsetC  = DIM*jnrC;
 592             j_coord_offsetD  = DIM*jnrD;
 593
 594             /* load j atom coordinates */
 595             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 596                                               x+j_coord_offsetC,x+j_coord_offsetD,
 597                                               &jx0,&jy0,&jz0);
 598
 599             /* Calculate displacement vector */
 600             dx00             = _mm_sub_ps(ix0,jx0);
 601             dy00             = _mm_sub_ps(iy0,jy0);
 602             dz00             = _mm_sub_ps(iz0,jz0);
 603
 604             /* Calculate squared distance and things based on it */
 605             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 606
 607             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 608
 609             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 610
 611             /* Load parameters for j particles */
 612             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 613                                                               charge+jnrC+0,charge+jnrD+0);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq00,rcutoff2))
 620             {
 621
 622             r00              = _mm_mul_ps(rsq00,rinv00);
 623             r00              = _mm_andnot_ps(dummy_mask,r00);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm_mul_ps(iq0,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Analytical PME correction */
 631             zeta2            = _mm_mul_ps(beta2,rsq00);
 632             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 633             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 634             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 635             felec            = _mm_mul_ps(qq00,felec);
 636
 637             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_ps(fscal,cutoff_mask);
 642
 643             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 644
 645              /* Update vectorial force */
 646             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 647             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 648             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 649
 650             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 651             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 652             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 653             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 654             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 655                                                    _mm_mul_ps(dx00,fscal),
 656                                                    _mm_mul_ps(dy00,fscal),
 657                                                    _mm_mul_ps(dz00,fscal));
 658
 659             }
 660
 661             /* Inner loop uses 32 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 7 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*32);
 681 }