src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse2_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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_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_sse2_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 SSE, 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           tx,ty,tz,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,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 114     ewtab            = fr->ic->tabq_coul_FDV0;
 115     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 116     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 117
 118     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 119     rcutoff_scalar   = fr->rcoulomb;
 120     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_ps();
 156         fiy0             = _mm_setzero_ps();
 157         fiz0             = _mm_setzero_ps();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               x+j_coord_offsetC,x+j_coord_offsetD,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_ps(ix0,jx0);
 186             dy00             = _mm_sub_ps(iy0,jy0);
 187             dz00             = _mm_sub_ps(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 193
 194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 198                                                               charge+jnrC+0,charge+jnrD+0);
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             if (gmx_mm_any_lt(rsq00,rcutoff2))
 205             {
 206
 207             r00              = _mm_mul_ps(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm_mul_ps(iq0,jq0);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _mm_mul_ps(r00,ewtabscale);
 216             ewitab           = _mm_cvttps_epi32(ewrt);
 217             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 218             ewitab           = _mm_slli_epi32(ewitab,2);
 219             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 220             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 221             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 222             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 223             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 224             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 225             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 226             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 227             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 228
 229             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velec            = _mm_and_ps(velec,cutoff_mask);
 233             velecsum         = _mm_add_ps(velecsum,velec);
 234
 235             fscal            = felec;
 236
 237             fscal            = _mm_and_ps(fscal,cutoff_mask);
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm_mul_ps(fscal,dx00);
 241             ty               = _mm_mul_ps(fscal,dy00);
 242             tz               = _mm_mul_ps(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_add_ps(fix0,tx);
 246             fiy0             = _mm_add_ps(fiy0,ty);
 247             fiz0             = _mm_add_ps(fiz0,tz);
 248
 249             fjptrA             = f+j_coord_offsetA;
 250             fjptrB             = f+j_coord_offsetB;
 251             fjptrC             = f+j_coord_offsetC;
 252             fjptrD             = f+j_coord_offsetD;
 253             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 254
 255             }
 256
 257             /* Inner loop uses 46 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             /* Get j neighbor index, and coordinate index */
 264             jnrlistA         = jjnr[jidx];
 265             jnrlistB         = jjnr[jidx+1];
 266             jnrlistC         = jjnr[jidx+2];
 267             jnrlistD         = jjnr[jidx+3];
 268             /* Sign of each element will be negative for non-real atoms.
 269              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 270              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 271              */
 272             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 273             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 274             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 275             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 276             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 277             j_coord_offsetA  = DIM*jnrA;
 278             j_coord_offsetB  = DIM*jnrB;
 279             j_coord_offsetC  = DIM*jnrC;
 280             j_coord_offsetD  = DIM*jnrD;
 281
 282             /* load j atom coordinates */
 283             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 284                                               x+j_coord_offsetC,x+j_coord_offsetD,
 285                                               &jx0,&jy0,&jz0);
 286
 287             /* Calculate displacement vector */
 288             dx00             = _mm_sub_ps(ix0,jx0);
 289             dy00             = _mm_sub_ps(iy0,jy0);
 290             dz00             = _mm_sub_ps(iz0,jz0);
 291
 292             /* Calculate squared distance and things based on it */
 293             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 294
 295             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 296
 297             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 298
 299             /* Load parameters for j particles */
 300             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 301                                                               charge+jnrC+0,charge+jnrD+0);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq00,rcutoff2))
 308             {
 309
 310             r00              = _mm_mul_ps(rsq00,rinv00);
 311             r00              = _mm_andnot_ps(dummy_mask,r00);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq00             = _mm_mul_ps(iq0,jq0);
 315
 316             /* EWALD ELECTROSTATICS */
 317
 318             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 319             ewrt             = _mm_mul_ps(r00,ewtabscale);
 320             ewitab           = _mm_cvttps_epi32(ewrt);
 321             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 322             ewitab           = _mm_slli_epi32(ewitab,2);
 323             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 324             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 325             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 326             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 327             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 328             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 329             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 330             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 331             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 332
 333             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_ps(velec,cutoff_mask);
 337             velec            = _mm_andnot_ps(dummy_mask,velec);
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_and_ps(fscal,cutoff_mask);
 343
 344             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx00);
 348             ty               = _mm_mul_ps(fscal,dy00);
 349             tz               = _mm_mul_ps(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm_add_ps(fix0,tx);
 353             fiy0             = _mm_add_ps(fiy0,ty);
 354             fiz0             = _mm_add_ps(fiz0,tz);
 355
 356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 360             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 361
 362             }
 363
 364             /* Inner loop uses 47 flops */
 365         }
 366
 367         /* End of innermost loop */
 368
 369         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 370                                               f+i_coord_offset,fshift+i_shift_offset);
 371
 372         ggid                        = gid[iidx];
 373         /* Update potential energies */
 374         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 8 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*47);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 391  * Electrostatics interaction: Ewald
 392  * VdW interaction:            None
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 398                     (t_nblist                    * gmx_restrict       nlist,
 399                      rvec                        * gmx_restrict          xx,
 400                      rvec                        * gmx_restrict          ff,
 401                      t_forcerec                  * gmx_restrict          fr,
 402                      t_mdatoms                   * gmx_restrict     mdatoms,
 403                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 404                      t_nrnb                      * gmx_restrict        nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB,jnrC,jnrD;
 414     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 415     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 417     real             rcutoff_scalar;
 418     real             *shiftvec,*fshift,*x,*f;
 419     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 420     real             scratch[4*DIM];
 421     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 425     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 428     real             *charge;
 429     __m128i          ewitab;
 430     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 431     real             *ewtab;
 432     __m128           dummy_mask,cutoff_mask;
 433     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 434     __m128           one     = _mm_set1_ps(1.0);
 435     __m128           two     = _mm_set1_ps(2.0);
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = _mm_set1_ps(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449
 450     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 451     ewtab            = fr->ic->tabq_coul_F;
 452     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 453     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 454
 455     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 456     rcutoff_scalar   = fr->rcoulomb;
 457     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 458     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 459
 460     /* Avoid stupid compiler warnings */
 461     jnrA = jnrB = jnrC = jnrD = 0;
 462     j_coord_offsetA = 0;
 463     j_coord_offsetB = 0;
 464     j_coord_offsetC = 0;
 465     j_coord_offsetD = 0;
 466
 467     outeriter        = 0;
 468     inneriter        = 0;
 469
 470     for(iidx=0;iidx<4*DIM;iidx++)
 471     {
 472         scratch[iidx] = 0.0;
 473     }
 474
 475     /* Start outer loop over neighborlists */
 476     for(iidx=0; iidx<nri; iidx++)
 477     {
 478         /* Load shift vector for this list */
 479         i_shift_offset   = DIM*shiftidx[iidx];
 480
 481         /* Load limits for loop over neighbors */
 482         j_index_start    = jindex[iidx];
 483         j_index_end      = jindex[iidx+1];
 484
 485         /* Get outer coordinate index */
 486         inr              = iinr[iidx];
 487         i_coord_offset   = DIM*inr;
 488
 489         /* Load i particle coords and add shift vector */
 490         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 491
 492         fix0             = _mm_setzero_ps();
 493         fiy0             = _mm_setzero_ps();
 494         fiz0             = _mm_setzero_ps();
 495
 496         /* Load parameters for i particles */
 497         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 498
 499         /* Start inner kernel loop */
 500         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrA             = jjnr[jidx];
 505             jnrB             = jjnr[jidx+1];
 506             jnrC             = jjnr[jidx+2];
 507             jnrD             = jjnr[jidx+3];
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510             j_coord_offsetC  = DIM*jnrC;
 511             j_coord_offsetD  = DIM*jnrD;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               x+j_coord_offsetC,x+j_coord_offsetD,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _mm_sub_ps(ix0,jx0);
 520             dy00             = _mm_sub_ps(iy0,jy0);
 521             dz00             = _mm_sub_ps(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 525
 526             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 527
 528             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 532                                                               charge+jnrC+0,charge+jnrD+0);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq00,rcutoff2))
 539             {
 540
 541             r00              = _mm_mul_ps(rsq00,rinv00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq00             = _mm_mul_ps(iq0,jq0);
 545
 546             /* EWALD ELECTROSTATICS */
 547
 548             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 549             ewrt             = _mm_mul_ps(r00,ewtabscale);
 550             ewitab           = _mm_cvttps_epi32(ewrt);
 551             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 552             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 553                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 554                                          &ewtabF,&ewtabFn);
 555             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 556             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 557
 558             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_and_ps(fscal,cutoff_mask);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx00);
 566             ty               = _mm_mul_ps(fscal,dy00);
 567             tz               = _mm_mul_ps(fscal,dz00);
 568
 569             /* Update vectorial force */
 570             fix0             = _mm_add_ps(fix0,tx);
 571             fiy0             = _mm_add_ps(fiy0,ty);
 572             fiz0             = _mm_add_ps(fiz0,tz);
 573
 574             fjptrA             = f+j_coord_offsetA;
 575             fjptrB             = f+j_coord_offsetB;
 576             fjptrC             = f+j_coord_offsetC;
 577             fjptrD             = f+j_coord_offsetD;
 578             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 579
 580             }
 581
 582             /* Inner loop uses 39 flops */
 583         }
 584
 585         if(jidx<j_index_end)
 586         {
 587
 588             /* Get j neighbor index, and coordinate index */
 589             jnrlistA         = jjnr[jidx];
 590             jnrlistB         = jjnr[jidx+1];
 591             jnrlistC         = jjnr[jidx+2];
 592             jnrlistD         = jjnr[jidx+3];
 593             /* Sign of each element will be negative for non-real atoms.
 594              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 595              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 596              */
 597             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 598             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 599             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 600             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 601             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 602             j_coord_offsetA  = DIM*jnrA;
 603             j_coord_offsetB  = DIM*jnrB;
 604             j_coord_offsetC  = DIM*jnrC;
 605             j_coord_offsetD  = DIM*jnrD;
 606
 607             /* load j atom coordinates */
 608             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 609                                               x+j_coord_offsetC,x+j_coord_offsetD,
 610                                               &jx0,&jy0,&jz0);
 611
 612             /* Calculate displacement vector */
 613             dx00             = _mm_sub_ps(ix0,jx0);
 614             dy00             = _mm_sub_ps(iy0,jy0);
 615             dz00             = _mm_sub_ps(iz0,jz0);
 616
 617             /* Calculate squared distance and things based on it */
 618             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 619
 620             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 621
 622             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 623
 624             /* Load parameters for j particles */
 625             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 626                                                               charge+jnrC+0,charge+jnrD+0);
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             if (gmx_mm_any_lt(rsq00,rcutoff2))
 633             {
 634
 635             r00              = _mm_mul_ps(rsq00,rinv00);
 636             r00              = _mm_andnot_ps(dummy_mask,r00);
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq00             = _mm_mul_ps(iq0,jq0);
 640
 641             /* EWALD ELECTROSTATICS */
 642
 643             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 644             ewrt             = _mm_mul_ps(r00,ewtabscale);
 645             ewitab           = _mm_cvttps_epi32(ewrt);
 646             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 647             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 648                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 649                                          &ewtabF,&ewtabFn);
 650             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 651             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 652
 653             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_ps(fscal,cutoff_mask);
 658
 659             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_ps(fscal,dx00);
 663             ty               = _mm_mul_ps(fscal,dy00);
 664             tz               = _mm_mul_ps(fscal,dz00);
 665
 666             /* Update vectorial force */
 667             fix0             = _mm_add_ps(fix0,tx);
 668             fiy0             = _mm_add_ps(fiy0,ty);
 669             fiz0             = _mm_add_ps(fiz0,tz);
 670
 671             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 672             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 673             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 674             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 676
 677             }
 678
 679             /* Inner loop uses 40 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 7 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 699 }