src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_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 SSE, 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           tx,ty,tz,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          vfitab;
  91     __m128i          ifour       = _mm_set1_epi32(4);
  92     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  93     real             *vftab;
  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     vftab            = kernel_data->table_elec->data;
 113     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 188                                                               charge+jnrC+0,charge+jnrD+0);
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _mm_mul_ps(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _mm_mul_ps(iq0,jq0);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _mm_mul_ps(r00,vftabscale);
 201             vfitab           = _mm_cvttps_epi32(rt);
 202             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 203             vfitab           = _mm_slli_epi32(vfitab,2);
 204
 205             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 206             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 207             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 208             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 209             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 210             _MM_TRANSPOSE4_PS(Y,F,G,H);
 211             Heps             = _mm_mul_ps(vfeps,H);
 212             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 213             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 214             velec            = _mm_mul_ps(qq00,VV);
 215             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 216             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 217
 218             /* Update potential sum for this i atom from the interaction with this j atom. */
 219             velecsum         = _mm_add_ps(velecsum,velec);
 220
 221             fscal            = felec;
 222
 223             /* Calculate temporary vectorial force */
 224             tx               = _mm_mul_ps(fscal,dx00);
 225             ty               = _mm_mul_ps(fscal,dy00);
 226             tz               = _mm_mul_ps(fscal,dz00);
 227
 228             /* Update vectorial force */
 229             fix0             = _mm_add_ps(fix0,tx);
 230             fiy0             = _mm_add_ps(fiy0,ty);
 231             fiz0             = _mm_add_ps(fiz0,tz);
 232
 233             fjptrA             = f+j_coord_offsetA;
 234             fjptrB             = f+j_coord_offsetB;
 235             fjptrC             = f+j_coord_offsetC;
 236             fjptrD             = f+j_coord_offsetD;
 237             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 238
 239             /* Inner loop uses 43 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             /* Get j neighbor index, and coordinate index */
 246             jnrlistA         = jjnr[jidx];
 247             jnrlistB         = jjnr[jidx+1];
 248             jnrlistC         = jjnr[jidx+2];
 249             jnrlistD         = jjnr[jidx+3];
 250             /* Sign of each element will be negative for non-real atoms.
 251              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 252              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 253              */
 254             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 255             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 256             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 257             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 258             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                               x+j_coord_offsetC,x+j_coord_offsetD,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm_sub_ps(ix0,jx0);
 271             dy00             = _mm_sub_ps(iy0,jy0);
 272             dz00             = _mm_sub_ps(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 278
 279             /* Load parameters for j particles */
 280             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 281                                                               charge+jnrC+0,charge+jnrD+0);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r00              = _mm_mul_ps(rsq00,rinv00);
 288             r00              = _mm_andnot_ps(dummy_mask,r00);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq00             = _mm_mul_ps(iq0,jq0);
 292
 293             /* Calculate table index by multiplying r with table scale and truncate to integer */
 294             rt               = _mm_mul_ps(r00,vftabscale);
 295             vfitab           = _mm_cvttps_epi32(rt);
 296             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 297             vfitab           = _mm_slli_epi32(vfitab,2);
 298
 299             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 300             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 302             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 303             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 304             _MM_TRANSPOSE4_PS(Y,F,G,H);
 305             Heps             = _mm_mul_ps(vfeps,H);
 306             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 307             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 308             velec            = _mm_mul_ps(qq00,VV);
 309             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 310             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_andnot_ps(dummy_mask,velec);
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_ps(fscal,dx00);
 322             ty               = _mm_mul_ps(fscal,dy00);
 323             tz               = _mm_mul_ps(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm_add_ps(fix0,tx);
 327             fiy0             = _mm_add_ps(fiy0,ty);
 328             fiz0             = _mm_add_ps(fiz0,tz);
 329
 330             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 331             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 332             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 333             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 335
 336             /* Inner loop uses 44 flops */
 337         }
 338
 339         /* End of innermost loop */
 340
 341         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 342                                               f+i_coord_offset,fshift+i_shift_offset);
 343
 344         ggid                        = gid[iidx];
 345         /* Update potential energies */
 346         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 347
 348         /* Increment number of inner iterations */
 349         inneriter                  += j_index_end - j_index_start;
 350
 351         /* Outer loop uses 8 flops */
 352     }
 353
 354     /* Increment number of outer iterations */
 355     outeriter        += nri;
 356
 357     /* Update outer/inner flops */
 358
 359     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 360 }
 361 /*
 362  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 363  * Electrostatics interaction: CubicSplineTable
 364  * VdW interaction:            None
 365  * Geometry:                   Particle-Particle
 366  * Calculate force/pot:        Force
 367  */
 368 void
 369 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 370                     (t_nblist                    * gmx_restrict       nlist,
 371                      rvec                        * gmx_restrict          xx,
 372                      rvec                        * gmx_restrict          ff,
 373                      t_forcerec                  * gmx_restrict          fr,
 374                      t_mdatoms                   * gmx_restrict     mdatoms,
 375                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 376                      t_nrnb                      * gmx_restrict        nrnb)
 377 {
 378     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 379      * just 0 for non-waters.
 380      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 381      * jnr indices corresponding to data put in the four positions in the SIMD register.
 382      */
 383     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 384     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 385     int              jnrA,jnrB,jnrC,jnrD;
 386     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 387     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 392     real             scratch[4*DIM];
 393     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     int              vdwioffset0;
 395     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 397     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     __m128i          vfitab;
 402     __m128i          ifour       = _mm_set1_epi32(4);
 403     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 404     real             *vftab;
 405     __m128           dummy_mask,cutoff_mask;
 406     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 407     __m128           one     = _mm_set1_ps(1.0);
 408     __m128           two     = _mm_set1_ps(2.0);
 409     x                = xx[0];
 410     f                = ff[0];
 411
 412     nri              = nlist->nri;
 413     iinr             = nlist->iinr;
 414     jindex           = nlist->jindex;
 415     jjnr             = nlist->jjnr;
 416     shiftidx         = nlist->shift;
 417     gid              = nlist->gid;
 418     shiftvec         = fr->shift_vec[0];
 419     fshift           = fr->fshift[0];
 420     facel            = _mm_set1_ps(fr->epsfac);
 421     charge           = mdatoms->chargeA;
 422
 423     vftab            = kernel_data->table_elec->data;
 424     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 425
 426     /* Avoid stupid compiler warnings */
 427     jnrA = jnrB = jnrC = jnrD = 0;
 428     j_coord_offsetA = 0;
 429     j_coord_offsetB = 0;
 430     j_coord_offsetC = 0;
 431     j_coord_offsetD = 0;
 432
 433     outeriter        = 0;
 434     inneriter        = 0;
 435
 436     for(iidx=0;iidx<4*DIM;iidx++)
 437     {
 438         scratch[iidx] = 0.0;
 439     }
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _mm_setzero_ps();
 459         fiy0             = _mm_setzero_ps();
 460         fiz0             = _mm_setzero_ps();
 461
 462         /* Load parameters for i particles */
 463         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             jnrC             = jjnr[jidx+2];
 473             jnrD             = jjnr[jidx+3];
 474             j_coord_offsetA  = DIM*jnrA;
 475             j_coord_offsetB  = DIM*jnrB;
 476             j_coord_offsetC  = DIM*jnrC;
 477             j_coord_offsetD  = DIM*jnrD;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                               x+j_coord_offsetC,x+j_coord_offsetD,
 482                                               &jx0,&jy0,&jz0);
 483
 484             /* Calculate displacement vector */
 485             dx00             = _mm_sub_ps(ix0,jx0);
 486             dy00             = _mm_sub_ps(iy0,jy0);
 487             dz00             = _mm_sub_ps(iz0,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 491
 492             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r00              = _mm_mul_ps(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq00             = _mm_mul_ps(iq0,jq0);
 506
 507             /* Calculate table index by multiplying r with table scale and truncate to integer */
 508             rt               = _mm_mul_ps(r00,vftabscale);
 509             vfitab           = _mm_cvttps_epi32(rt);
 510             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 511             vfitab           = _mm_slli_epi32(vfitab,2);
 512
 513             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 514             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 515             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 516             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 517             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 518             _MM_TRANSPOSE4_PS(Y,F,G,H);
 519             Heps             = _mm_mul_ps(vfeps,H);
 520             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 521             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 522             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 523
 524             fscal            = felec;
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm_mul_ps(fscal,dx00);
 528             ty               = _mm_mul_ps(fscal,dy00);
 529             tz               = _mm_mul_ps(fscal,dz00);
 530
 531             /* Update vectorial force */
 532             fix0             = _mm_add_ps(fix0,tx);
 533             fiy0             = _mm_add_ps(fiy0,ty);
 534             fiz0             = _mm_add_ps(fiz0,tz);
 535
 536             fjptrA             = f+j_coord_offsetA;
 537             fjptrB             = f+j_coord_offsetB;
 538             fjptrC             = f+j_coord_offsetC;
 539             fjptrD             = f+j_coord_offsetD;
 540             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 541
 542             /* Inner loop uses 39 flops */
 543         }
 544
 545         if(jidx<j_index_end)
 546         {
 547
 548             /* Get j neighbor index, and coordinate index */
 549             jnrlistA         = jjnr[jidx];
 550             jnrlistB         = jjnr[jidx+1];
 551             jnrlistC         = jjnr[jidx+2];
 552             jnrlistD         = jjnr[jidx+3];
 553             /* Sign of each element will be negative for non-real atoms.
 554              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 555              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 556              */
 557             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 558             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 559             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 560             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 561             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 562             j_coord_offsetA  = DIM*jnrA;
 563             j_coord_offsetB  = DIM*jnrB;
 564             j_coord_offsetC  = DIM*jnrC;
 565             j_coord_offsetD  = DIM*jnrD;
 566
 567             /* load j atom coordinates */
 568             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 569                                               x+j_coord_offsetC,x+j_coord_offsetD,
 570                                               &jx0,&jy0,&jz0);
 571
 572             /* Calculate displacement vector */
 573             dx00             = _mm_sub_ps(ix0,jx0);
 574             dy00             = _mm_sub_ps(iy0,jy0);
 575             dz00             = _mm_sub_ps(iz0,jz0);
 576
 577             /* Calculate squared distance and things based on it */
 578             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 579
 580             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 581
 582             /* Load parameters for j particles */
 583             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 584                                                               charge+jnrC+0,charge+jnrD+0);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r00              = _mm_mul_ps(rsq00,rinv00);
 591             r00              = _mm_andnot_ps(dummy_mask,r00);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq00             = _mm_mul_ps(iq0,jq0);
 595
 596             /* Calculate table index by multiplying r with table scale and truncate to integer */
 597             rt               = _mm_mul_ps(r00,vftabscale);
 598             vfitab           = _mm_cvttps_epi32(rt);
 599             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 600             vfitab           = _mm_slli_epi32(vfitab,2);
 601
 602             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 603             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 604             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 605             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 606             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 607             _MM_TRANSPOSE4_PS(Y,F,G,H);
 608             Heps             = _mm_mul_ps(vfeps,H);
 609             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 610             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 611             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_ps(fscal,dx00);
 619             ty               = _mm_mul_ps(fscal,dy00);
 620             tz               = _mm_mul_ps(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_add_ps(fix0,tx);
 624             fiy0             = _mm_add_ps(fiy0,ty);
 625             fiz0             = _mm_add_ps(fiz0,tz);
 626
 627             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 628             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 629             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 630             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 631             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 632
 633             /* Inner loop uses 40 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 7 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 653 }