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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_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          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113
 114     vftab            = kernel_data->table_elec->data;
 115     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 190                                                               charge+jnrC+0,charge+jnrD+0);
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_ps(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_ps(iq0,jq0);
 200
 201             /* Calculate table index by multiplying r with table scale and truncate to integer */
 202             rt               = _mm_mul_ps(r00,vftabscale);
 203             vfitab           = _mm_cvttps_epi32(rt);
 204             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 205             vfitab           = _mm_slli_epi32(vfitab,2);
 206
 207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 208             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 209             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 210             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 211             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 212             _MM_TRANSPOSE4_PS(Y,F,G,H);
 213             Heps             = _mm_mul_ps(vfeps,H);
 214             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 215             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 216             velec            = _mm_mul_ps(qq00,VV);
 217             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 218             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             velecsum         = _mm_add_ps(velecsum,velec);
 222
 223             fscal            = felec;
 224
 225             /* Calculate temporary vectorial force */
 226             tx               = _mm_mul_ps(fscal,dx00);
 227             ty               = _mm_mul_ps(fscal,dy00);
 228             tz               = _mm_mul_ps(fscal,dz00);
 229
 230             /* Update vectorial force */
 231             fix0             = _mm_add_ps(fix0,tx);
 232             fiy0             = _mm_add_ps(fiy0,ty);
 233             fiz0             = _mm_add_ps(fiz0,tz);
 234
 235             fjptrA             = f+j_coord_offsetA;
 236             fjptrB             = f+j_coord_offsetB;
 237             fjptrC             = f+j_coord_offsetC;
 238             fjptrD             = f+j_coord_offsetD;
 239             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 240
 241             /* Inner loop uses 43 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             /* Get j neighbor index, and coordinate index */
 248             jnrlistA         = jjnr[jidx];
 249             jnrlistB         = jjnr[jidx+1];
 250             jnrlistC         = jjnr[jidx+2];
 251             jnrlistD         = jjnr[jidx+3];
 252             /* Sign of each element will be negative for non-real atoms.
 253              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 254              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 255              */
 256             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 257             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 258             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 259             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 260             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 261             j_coord_offsetA  = DIM*jnrA;
 262             j_coord_offsetB  = DIM*jnrB;
 263             j_coord_offsetC  = DIM*jnrC;
 264             j_coord_offsetD  = DIM*jnrD;
 265
 266             /* load j atom coordinates */
 267             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 268                                               x+j_coord_offsetC,x+j_coord_offsetD,
 269                                               &jx0,&jy0,&jz0);
 270
 271             /* Calculate displacement vector */
 272             dx00             = _mm_sub_ps(ix0,jx0);
 273             dy00             = _mm_sub_ps(iy0,jy0);
 274             dz00             = _mm_sub_ps(iz0,jz0);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 278
 279             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 283                                                               charge+jnrC+0,charge+jnrD+0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r00              = _mm_mul_ps(rsq00,rinv00);
 290             r00              = _mm_andnot_ps(dummy_mask,r00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm_mul_ps(iq0,jq0);
 294
 295             /* Calculate table index by multiplying r with table scale and truncate to integer */
 296             rt               = _mm_mul_ps(r00,vftabscale);
 297             vfitab           = _mm_cvttps_epi32(rt);
 298             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 299             vfitab           = _mm_slli_epi32(vfitab,2);
 300
 301             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 302             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 303             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 304             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 305             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 306             _MM_TRANSPOSE4_PS(Y,F,G,H);
 307             Heps             = _mm_mul_ps(vfeps,H);
 308             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 309             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 310             velec            = _mm_mul_ps(qq00,VV);
 311             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 312             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_andnot_ps(dummy_mask,velec);
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_ps(fscal,dx00);
 324             ty               = _mm_mul_ps(fscal,dy00);
 325             tz               = _mm_mul_ps(fscal,dz00);
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_add_ps(fix0,tx);
 329             fiy0             = _mm_add_ps(fiy0,ty);
 330             fiz0             = _mm_add_ps(fiz0,tz);
 331
 332             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 333             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 334             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 335             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 336             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 337
 338             /* Inner loop uses 44 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 8 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 365  * Electrostatics interaction: CubicSplineTable
 366  * VdW interaction:            None
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      t_forcerec                  * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB,jnrC,jnrD;
 388     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 389     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 390     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 391     real             rcutoff_scalar;
 392     real             *shiftvec,*fshift,*x,*f;
 393     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 394     real             scratch[4*DIM];
 395     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 396     int              vdwioffset0;
 397     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 398     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 399     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 400     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 401     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 402     real             *charge;
 403     __m128i          vfitab;
 404     __m128i          ifour       = _mm_set1_epi32(4);
 405     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 406     real             *vftab;
 407     __m128           dummy_mask,cutoff_mask;
 408     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 409     __m128           one     = _mm_set1_ps(1.0);
 410     __m128           two     = _mm_set1_ps(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_ps(fr->epsfac);
 423     charge           = mdatoms->chargeA;
 424
 425     vftab            = kernel_data->table_elec->data;
 426     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = jnrC = jnrD = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432     j_coord_offsetC = 0;
 433     j_coord_offsetD = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     for(iidx=0;iidx<4*DIM;iidx++)
 439     {
 440         scratch[iidx] = 0.0;
 441     }
 442
 443     /* Start outer loop over neighborlists */
 444     for(iidx=0; iidx<nri; iidx++)
 445     {
 446         /* Load shift vector for this list */
 447         i_shift_offset   = DIM*shiftidx[iidx];
 448
 449         /* Load limits for loop over neighbors */
 450         j_index_start    = jindex[iidx];
 451         j_index_end      = jindex[iidx+1];
 452
 453         /* Get outer coordinate index */
 454         inr              = iinr[iidx];
 455         i_coord_offset   = DIM*inr;
 456
 457         /* Load i particle coords and add shift vector */
 458         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 459
 460         fix0             = _mm_setzero_ps();
 461         fiy0             = _mm_setzero_ps();
 462         fiz0             = _mm_setzero_ps();
 463
 464         /* Load parameters for i particles */
 465         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             jnrC             = jjnr[jidx+2];
 475             jnrD             = jjnr[jidx+3];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               x+j_coord_offsetC,x+j_coord_offsetD,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_ps(ix0,jx0);
 488             dy00             = _mm_sub_ps(iy0,jy0);
 489             dz00             = _mm_sub_ps(iz0,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 493
 494             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                               charge+jnrC+0,charge+jnrD+0);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             r00              = _mm_mul_ps(rsq00,rinv00);
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             qq00             = _mm_mul_ps(iq0,jq0);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm_mul_ps(r00,vftabscale);
 511             vfitab           = _mm_cvttps_epi32(rt);
 512             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 513             vfitab           = _mm_slli_epi32(vfitab,2);
 514
 515             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 516             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 517             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 518             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 519             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 520             _MM_TRANSPOSE4_PS(Y,F,G,H);
 521             Heps             = _mm_mul_ps(vfeps,H);
 522             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 523             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 524             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 525
 526             fscal            = felec;
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx00);
 530             ty               = _mm_mul_ps(fscal,dy00);
 531             tz               = _mm_mul_ps(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_ps(fix0,tx);
 535             fiy0             = _mm_add_ps(fiy0,ty);
 536             fiz0             = _mm_add_ps(fiz0,tz);
 537
 538             fjptrA             = f+j_coord_offsetA;
 539             fjptrB             = f+j_coord_offsetB;
 540             fjptrC             = f+j_coord_offsetC;
 541             fjptrD             = f+j_coord_offsetD;
 542             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 543
 544             /* Inner loop uses 39 flops */
 545         }
 546
 547         if(jidx<j_index_end)
 548         {
 549
 550             /* Get j neighbor index, and coordinate index */
 551             jnrlistA         = jjnr[jidx];
 552             jnrlistB         = jjnr[jidx+1];
 553             jnrlistC         = jjnr[jidx+2];
 554             jnrlistD         = jjnr[jidx+3];
 555             /* Sign of each element will be negative for non-real atoms.
 556              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 557              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 558              */
 559             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 560             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 561             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 562             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 563             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 564             j_coord_offsetA  = DIM*jnrA;
 565             j_coord_offsetB  = DIM*jnrB;
 566             j_coord_offsetC  = DIM*jnrC;
 567             j_coord_offsetD  = DIM*jnrD;
 568
 569             /* load j atom coordinates */
 570             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 571                                               x+j_coord_offsetC,x+j_coord_offsetD,
 572                                               &jx0,&jy0,&jz0);
 573
 574             /* Calculate displacement vector */
 575             dx00             = _mm_sub_ps(ix0,jx0);
 576             dy00             = _mm_sub_ps(iy0,jy0);
 577             dz00             = _mm_sub_ps(iz0,jz0);
 578
 579             /* Calculate squared distance and things based on it */
 580             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 581
 582             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 583
 584             /* Load parameters for j particles */
 585             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 586                                                               charge+jnrC+0,charge+jnrD+0);
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             r00              = _mm_mul_ps(rsq00,rinv00);
 593             r00              = _mm_andnot_ps(dummy_mask,r00);
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq00             = _mm_mul_ps(iq0,jq0);
 597
 598             /* Calculate table index by multiplying r with table scale and truncate to integer */
 599             rt               = _mm_mul_ps(r00,vftabscale);
 600             vfitab           = _mm_cvttps_epi32(rt);
 601             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 602             vfitab           = _mm_slli_epi32(vfitab,2);
 603
 604             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 605             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 606             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 607             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 608             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 609             _MM_TRANSPOSE4_PS(Y,F,G,H);
 610             Heps             = _mm_mul_ps(vfeps,H);
 611             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 612             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 613             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx00);
 621             ty               = _mm_mul_ps(fscal,dy00);
 622             tz               = _mm_mul_ps(fscal,dz00);
 623
 624             /* Update vectorial force */
 625             fix0             = _mm_add_ps(fix0,tx);
 626             fiy0             = _mm_add_ps(fiy0,ty);
 627             fiz0             = _mm_add_ps(fiz0,tz);
 628
 629             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 630             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 631             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 632             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 633             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 634
 635             /* Inner loop uses 40 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         /* Increment number of inner iterations */
 644         inneriter                  += j_index_end - j_index_start;
 645
 646         /* Outer loop uses 7 flops */
 647     }
 648
 649     /* Increment number of outer iterations */
 650     outeriter        += nri;
 651
 652     /* Update outer/inner flops */
 653
 654     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 655 }