src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 114
 115     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->ic->rcoulomb;
 117     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 118     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               x+j_coord_offsetC,x+j_coord_offsetD,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_ps(ix0,jx0);
 183             dy00             = _mm_sub_ps(iy0,jy0);
 184             dz00             = _mm_sub_ps(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 188
 189             rinv00           = sse2_invsqrt_f(rsq00);
 190
 191             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 195                                                               charge+jnrC+0,charge+jnrD+0);
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             r00              = _mm_mul_ps(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_ps(iq0,jq0);
 208
 209             /* EWALD ELECTROSTATICS */
 210
 211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 212             ewrt             = _mm_mul_ps(r00,ewtabscale);
 213             ewitab           = _mm_cvttps_epi32(ewrt);
 214             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 215             ewitab           = _mm_slli_epi32(ewitab,2);
 216             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 217             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 218             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 219             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 220             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 221             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 222             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 223             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 224             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 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             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_ps(fscal,dx00);
 238             ty               = _mm_mul_ps(fscal,dy00);
 239             tz               = _mm_mul_ps(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_ps(fix0,tx);
 243             fiy0             = _mm_add_ps(fiy0,ty);
 244             fiz0             = _mm_add_ps(fiz0,tz);
 245
 246             fjptrA             = f+j_coord_offsetA;
 247             fjptrB             = f+j_coord_offsetB;
 248             fjptrC             = f+j_coord_offsetC;
 249             fjptrD             = f+j_coord_offsetD;
 250             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 251
 252             }
 253
 254             /* Inner loop uses 46 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           = sse2_invsqrt_f(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             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 316             ewrt             = _mm_mul_ps(r00,ewtabscale);
 317             ewitab           = _mm_cvttps_epi32(ewrt);
 318             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 319             ewitab           = _mm_slli_epi32(ewitab,2);
 320             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 321             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 322             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 323             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 324             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 325             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 326             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 327             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 328             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 329
 330             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_ps(velec,cutoff_mask);
 334             velec            = _mm_andnot_ps(dummy_mask,velec);
 335             velecsum         = _mm_add_ps(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_and_ps(fscal,cutoff_mask);
 340
 341             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_ps(fscal,dx00);
 345             ty               = _mm_mul_ps(fscal,dy00);
 346             tz               = _mm_mul_ps(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm_add_ps(fix0,tx);
 350             fiy0             = _mm_add_ps(fiy0,ty);
 351             fiz0             = _mm_add_ps(fiz0,tz);
 352
 353             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 354             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 355             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 356             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 357             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 358
 359             }
 360
 361             /* Inner loop uses 47 flops */
 362         }
 363
 364         /* End of innermost loop */
 365
 366         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 367                                               f+i_coord_offset,fshift+i_shift_offset);
 368
 369         ggid                        = gid[iidx];
 370         /* Update potential energies */
 371         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 372
 373         /* Increment number of inner iterations */
 374         inneriter                  += j_index_end - j_index_start;
 375
 376         /* Outer loop uses 8 flops */
 377     }
 378
 379     /* Increment number of outer iterations */
 380     outeriter        += nri;
 381
 382     /* Update outer/inner flops */
 383
 384     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*47);
 385 }
 386 /*
 387  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 388  * Electrostatics interaction: Ewald
 389  * VdW interaction:            None
 390  * Geometry:                   Particle-Particle
 391  * Calculate force/pot:        Force
 392  */
 393 void
 394 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 395                     (t_nblist                    * gmx_restrict       nlist,
 396                      rvec                        * gmx_restrict          xx,
 397                      rvec                        * gmx_restrict          ff,
 398                      struct t_forcerec           * gmx_restrict          fr,
 399                      t_mdatoms                   * gmx_restrict     mdatoms,
 400                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 401                      t_nrnb                      * gmx_restrict        nrnb)
 402 {
 403     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 404      * just 0 for non-waters.
 405      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 406      * jnr indices corresponding to data put in the four positions in the SIMD register.
 407      */
 408     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 409     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 410     int              jnrA,jnrB,jnrC,jnrD;
 411     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 412     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 413     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 414     real             rcutoff_scalar;
 415     real             *shiftvec,*fshift,*x,*f;
 416     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 417     real             scratch[4*DIM];
 418     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 422     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     __m128i          ewitab;
 427     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 428     real             *ewtab;
 429     __m128           dummy_mask,cutoff_mask;
 430     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 431     __m128           one     = _mm_set1_ps(1.0);
 432     __m128           two     = _mm_set1_ps(2.0);
 433     x                = xx[0];
 434     f                = ff[0];
 435
 436     nri              = nlist->nri;
 437     iinr             = nlist->iinr;
 438     jindex           = nlist->jindex;
 439     jjnr             = nlist->jjnr;
 440     shiftidx         = nlist->shift;
 441     gid              = nlist->gid;
 442     shiftvec         = fr->shift_vec[0];
 443     fshift           = fr->fshift[0];
 444     facel            = _mm_set1_ps(fr->ic->epsfac);
 445     charge           = mdatoms->chargeA;
 446
 447     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 448     ewtab            = fr->ic->tabq_coul_F;
 449     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 450     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 451
 452     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 453     rcutoff_scalar   = fr->ic->rcoulomb;
 454     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 455     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 456
 457     /* Avoid stupid compiler warnings */
 458     jnrA = jnrB = jnrC = jnrD = 0;
 459     j_coord_offsetA = 0;
 460     j_coord_offsetB = 0;
 461     j_coord_offsetC = 0;
 462     j_coord_offsetD = 0;
 463
 464     outeriter        = 0;
 465     inneriter        = 0;
 466
 467     for(iidx=0;iidx<4*DIM;iidx++)
 468     {
 469         scratch[iidx] = 0.0;
 470     }
 471
 472     /* Start outer loop over neighborlists */
 473     for(iidx=0; iidx<nri; iidx++)
 474     {
 475         /* Load shift vector for this list */
 476         i_shift_offset   = DIM*shiftidx[iidx];
 477
 478         /* Load limits for loop over neighbors */
 479         j_index_start    = jindex[iidx];
 480         j_index_end      = jindex[iidx+1];
 481
 482         /* Get outer coordinate index */
 483         inr              = iinr[iidx];
 484         i_coord_offset   = DIM*inr;
 485
 486         /* Load i particle coords and add shift vector */
 487         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 488
 489         fix0             = _mm_setzero_ps();
 490         fiy0             = _mm_setzero_ps();
 491         fiz0             = _mm_setzero_ps();
 492
 493         /* Load parameters for i particles */
 494         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 495
 496         /* Start inner kernel loop */
 497         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 498         {
 499
 500             /* Get j neighbor index, and coordinate index */
 501             jnrA             = jjnr[jidx];
 502             jnrB             = jjnr[jidx+1];
 503             jnrC             = jjnr[jidx+2];
 504             jnrD             = jjnr[jidx+3];
 505             j_coord_offsetA  = DIM*jnrA;
 506             j_coord_offsetB  = DIM*jnrB;
 507             j_coord_offsetC  = DIM*jnrC;
 508             j_coord_offsetD  = DIM*jnrD;
 509
 510             /* load j atom coordinates */
 511             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                               x+j_coord_offsetC,x+j_coord_offsetD,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm_sub_ps(ix0,jx0);
 517             dy00             = _mm_sub_ps(iy0,jy0);
 518             dz00             = _mm_sub_ps(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 522
 523             rinv00           = sse2_invsqrt_f(rsq00);
 524
 525             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 529                                                               charge+jnrC+0,charge+jnrD+0);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             if (gmx_mm_any_lt(rsq00,rcutoff2))
 536             {
 537
 538             r00              = _mm_mul_ps(rsq00,rinv00);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq00             = _mm_mul_ps(iq0,jq0);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 546             ewrt             = _mm_mul_ps(r00,ewtabscale);
 547             ewitab           = _mm_cvttps_epi32(ewrt);
 548             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 549             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 550                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 551                                          &ewtabF,&ewtabFn);
 552             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 553             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 554
 555             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_and_ps(fscal,cutoff_mask);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx00);
 563             ty               = _mm_mul_ps(fscal,dy00);
 564             tz               = _mm_mul_ps(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm_add_ps(fix0,tx);
 568             fiy0             = _mm_add_ps(fiy0,ty);
 569             fiz0             = _mm_add_ps(fiz0,tz);
 570
 571             fjptrA             = f+j_coord_offsetA;
 572             fjptrB             = f+j_coord_offsetB;
 573             fjptrC             = f+j_coord_offsetC;
 574             fjptrD             = f+j_coord_offsetD;
 575             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 576
 577             }
 578
 579             /* Inner loop uses 39 flops */
 580         }
 581
 582         if(jidx<j_index_end)
 583         {
 584
 585             /* Get j neighbor index, and coordinate index */
 586             jnrlistA         = jjnr[jidx];
 587             jnrlistB         = jjnr[jidx+1];
 588             jnrlistC         = jjnr[jidx+2];
 589             jnrlistD         = jjnr[jidx+3];
 590             /* Sign of each element will be negative for non-real atoms.
 591              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 592              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 593              */
 594             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 595             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 596             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 597             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 598             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 599             j_coord_offsetA  = DIM*jnrA;
 600             j_coord_offsetB  = DIM*jnrB;
 601             j_coord_offsetC  = DIM*jnrC;
 602             j_coord_offsetD  = DIM*jnrD;
 603
 604             /* load j atom coordinates */
 605             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 606                                               x+j_coord_offsetC,x+j_coord_offsetD,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _mm_sub_ps(ix0,jx0);
 611             dy00             = _mm_sub_ps(iy0,jy0);
 612             dz00             = _mm_sub_ps(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 616
 617             rinv00           = sse2_invsqrt_f(rsq00);
 618
 619             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 620
 621             /* Load parameters for j particles */
 622             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 623                                                               charge+jnrC+0,charge+jnrD+0);
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             if (gmx_mm_any_lt(rsq00,rcutoff2))
 630             {
 631
 632             r00              = _mm_mul_ps(rsq00,rinv00);
 633             r00              = _mm_andnot_ps(dummy_mask,r00);
 634
 635             /* Compute parameters for interactions between i and j atoms */
 636             qq00             = _mm_mul_ps(iq0,jq0);
 637
 638             /* EWALD ELECTROSTATICS */
 639
 640             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 641             ewrt             = _mm_mul_ps(r00,ewtabscale);
 642             ewitab           = _mm_cvttps_epi32(ewrt);
 643             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 644             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 645                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 646                                          &ewtabF,&ewtabFn);
 647             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 648             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 649
 650             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_and_ps(fscal,cutoff_mask);
 655
 656             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm_mul_ps(fscal,dx00);
 660             ty               = _mm_mul_ps(fscal,dy00);
 661             tz               = _mm_mul_ps(fscal,dz00);
 662
 663             /* Update vectorial force */
 664             fix0             = _mm_add_ps(fix0,tx);
 665             fiy0             = _mm_add_ps(fiy0,ty);
 666             fiz0             = _mm_add_ps(fiz0,tz);
 667
 668             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 669             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 670             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 671             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 672             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 673
 674             }
 675
 676             /* Inner loop uses 40 flops */
 677         }
 678
 679         /* End of innermost loop */
 680
 681         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 682                                               f+i_coord_offset,fshift+i_shift_offset);
 683
 684         /* Increment number of inner iterations */
 685         inneriter                  += j_index_end - j_index_start;
 686
 687         /* Outer loop uses 7 flops */
 688     }
 689
 690     /* Increment number of outer iterations */
 691     outeriter        += nri;
 692
 693     /* Update outer/inner flops */
 694
 695     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 696 }