src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     int              nvdwtype;
  77     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  78     int              *vdwtype;
  79     real             *vdwparam;
  80     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  81     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = jnrC = jnrD = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107     j_coord_offsetC = 0;
 108     j_coord_offsetD = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     for(iidx=0;iidx<4*DIM;iidx++)
 114     {
 115         scratch[iidx] = 0.0;
 116     }
 117
 118     /* Start outer loop over neighborlists */
 119     for(iidx=0; iidx<nri; iidx++)
 120     {
 121         /* Load shift vector for this list */
 122         i_shift_offset   = DIM*shiftidx[iidx];
 123
 124         /* Load limits for loop over neighbors */
 125         j_index_start    = jindex[iidx];
 126         j_index_end      = jindex[iidx+1];
 127
 128         /* Get outer coordinate index */
 129         inr              = iinr[iidx];
 130         i_coord_offset   = DIM*inr;
 131
 132         /* Load i particle coords and add shift vector */
 133         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 134
 135         fix0             = _mm_setzero_ps();
 136         fiy0             = _mm_setzero_ps();
 137         fiz0             = _mm_setzero_ps();
 138
 139         /* Load parameters for i particles */
 140         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 141         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_ps();
 145         vvdwsum          = _mm_setzero_ps();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             jnrC             = jjnr[jidx+2];
 155             jnrD             = jjnr[jidx+3];
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158             j_coord_offsetC  = DIM*jnrC;
 159             j_coord_offsetD  = DIM*jnrD;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               x+j_coord_offsetC,x+j_coord_offsetD,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_ps(ix0,jx0);
 168             dy00             = _mm_sub_ps(iy0,jy0);
 169             dz00             = _mm_sub_ps(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 175
 176             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 180                                                               charge+jnrC+0,charge+jnrD+0);
 181             vdwjidx0A        = 2*vdwtype[jnrA+0];
 182             vdwjidx0B        = 2*vdwtype[jnrB+0];
 183             vdwjidx0C        = 2*vdwtype[jnrC+0];
 184             vdwjidx0D        = 2*vdwtype[jnrD+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             qq00             = _mm_mul_ps(iq0,jq0);
 192             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 193                                          vdwparam+vdwioffset0+vdwjidx0B,
 194                                          vdwparam+vdwioffset0+vdwjidx0C,
 195                                          vdwparam+vdwioffset0+vdwjidx0D,
 196                                          &c6_00,&c12_00);
 197
 198             /* COULOMB ELECTROSTATICS */
 199             velec            = _mm_mul_ps(qq00,rinv00);
 200             felec            = _mm_mul_ps(velec,rinvsq00);
 201
 202             /* LENNARD-JONES DISPERSION/REPULSION */
 203
 204             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 205             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 206             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 207             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 208             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             velecsum         = _mm_add_ps(velecsum,velec);
 212             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 213
 214             fscal            = _mm_add_ps(felec,fvdw);
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_ps(fscal,dx00);
 218             ty               = _mm_mul_ps(fscal,dy00);
 219             tz               = _mm_mul_ps(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_ps(fix0,tx);
 223             fiy0             = _mm_add_ps(fiy0,ty);
 224             fiz0             = _mm_add_ps(fiz0,tz);
 225
 226             fjptrA             = f+j_coord_offsetA;
 227             fjptrB             = f+j_coord_offsetB;
 228             fjptrC             = f+j_coord_offsetC;
 229             fjptrD             = f+j_coord_offsetD;
 230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 231
 232             /* Inner loop uses 40 flops */
 233         }
 234
 235         if(jidx<j_index_end)
 236         {
 237
 238             /* Get j neighbor index, and coordinate index */
 239             jnrlistA         = jjnr[jidx];
 240             jnrlistB         = jjnr[jidx+1];
 241             jnrlistC         = jjnr[jidx+2];
 242             jnrlistD         = jjnr[jidx+3];
 243             /* Sign of each element will be negative for non-real atoms.
 244              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 245              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 246              */
 247             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 248             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 249             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 250             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 251             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 252             j_coord_offsetA  = DIM*jnrA;
 253             j_coord_offsetB  = DIM*jnrB;
 254             j_coord_offsetC  = DIM*jnrC;
 255             j_coord_offsetD  = DIM*jnrD;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 259                                               x+j_coord_offsetC,x+j_coord_offsetD,
 260                                               &jx0,&jy0,&jz0);
 261
 262             /* Calculate displacement vector */
 263             dx00             = _mm_sub_ps(ix0,jx0);
 264             dy00             = _mm_sub_ps(iy0,jy0);
 265             dz00             = _mm_sub_ps(iz0,jz0);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 269
 270             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 271
 272             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 273
 274             /* Load parameters for j particles */
 275             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 276                                                               charge+jnrC+0,charge+jnrD+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq00             = _mm_mul_ps(iq0,jq0);
 288             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 289                                          vdwparam+vdwioffset0+vdwjidx0B,
 290                                          vdwparam+vdwioffset0+vdwjidx0C,
 291                                          vdwparam+vdwioffset0+vdwjidx0D,
 292                                          &c6_00,&c12_00);
 293
 294             /* COULOMB ELECTROSTATICS */
 295             velec            = _mm_mul_ps(qq00,rinv00);
 296             felec            = _mm_mul_ps(velec,rinvsq00);
 297
 298             /* LENNARD-JONES DISPERSION/REPULSION */
 299
 300             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 301             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 302             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 303             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 304             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm_andnot_ps(dummy_mask,velec);
 308             velecsum         = _mm_add_ps(velecsum,velec);
 309             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 310             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 311
 312             fscal            = _mm_add_ps(felec,fvdw);
 313
 314             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx00);
 318             ty               = _mm_mul_ps(fscal,dy00);
 319             tz               = _mm_mul_ps(fscal,dz00);
 320
 321             /* Update vectorial force */
 322             fix0             = _mm_add_ps(fix0,tx);
 323             fiy0             = _mm_add_ps(fiy0,ty);
 324             fiz0             = _mm_add_ps(fiz0,tz);
 325
 326             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 327             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 328             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 329             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 330             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 331
 332             /* Inner loop uses 40 flops */
 333         }
 334
 335         /* End of innermost loop */
 336
 337         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 338                                               f+i_coord_offset,fshift+i_shift_offset);
 339
 340         ggid                        = gid[iidx];
 341         /* Update potential energies */
 342         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 343         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 344
 345         /* Increment number of inner iterations */
 346         inneriter                  += j_index_end - j_index_start;
 347
 348         /* Outer loop uses 9 flops */
 349     }
 350
 351     /* Increment number of outer iterations */
 352     outeriter        += nri;
 353
 354     /* Update outer/inner flops */
 355
 356     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*40);
 357 }
 358 /*
 359  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 360  * Electrostatics interaction: Coulomb
 361  * VdW interaction:            LennardJones
 362  * Geometry:                   Particle-Particle
 363  * Calculate force/pot:        Force
 364  */
 365 void
 366 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 367                     (t_nblist * gmx_restrict                nlist,
 368                      rvec * gmx_restrict                    xx,
 369                      rvec * gmx_restrict                    ff,
 370                      t_forcerec * gmx_restrict              fr,
 371                      t_mdatoms * gmx_restrict               mdatoms,
 372                      nb_kernel_data_t * gmx_restrict        kernel_data,
 373                      t_nrnb * gmx_restrict                  nrnb)
 374 {
 375     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 376      * just 0 for non-waters.
 377      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 378      * jnr indices corresponding to data put in the four positions in the SIMD register.
 379      */
 380     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 381     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 382     int              jnrA,jnrB,jnrC,jnrD;
 383     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 384     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 385     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 386     real             rcutoff_scalar;
 387     real             *shiftvec,*fshift,*x,*f;
 388     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 389     real             scratch[4*DIM];
 390     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 391     int              vdwioffset0;
 392     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 393     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 394     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 395     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 396     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 397     real             *charge;
 398     int              nvdwtype;
 399     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 400     int              *vdwtype;
 401     real             *vdwparam;
 402     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 403     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 404     __m128           dummy_mask,cutoff_mask;
 405     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 406     __m128           one     = _mm_set1_ps(1.0);
 407     __m128           two     = _mm_set1_ps(2.0);
 408     x                = xx[0];
 409     f                = ff[0];
 410
 411     nri              = nlist->nri;
 412     iinr             = nlist->iinr;
 413     jindex           = nlist->jindex;
 414     jjnr             = nlist->jjnr;
 415     shiftidx         = nlist->shift;
 416     gid              = nlist->gid;
 417     shiftvec         = fr->shift_vec[0];
 418     fshift           = fr->fshift[0];
 419     facel            = _mm_set1_ps(fr->epsfac);
 420     charge           = mdatoms->chargeA;
 421     nvdwtype         = fr->ntype;
 422     vdwparam         = fr->nbfp;
 423     vdwtype          = mdatoms->typeA;
 424
 425     /* Avoid stupid compiler warnings */
 426     jnrA = jnrB = jnrC = jnrD = 0;
 427     j_coord_offsetA = 0;
 428     j_coord_offsetB = 0;
 429     j_coord_offsetC = 0;
 430     j_coord_offsetD = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     for(iidx=0;iidx<4*DIM;iidx++)
 436     {
 437         scratch[iidx] = 0.0;
 438     }
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_ps();
 458         fiy0             = _mm_setzero_ps();
 459         fiz0             = _mm_setzero_ps();
 460
 461         /* Load parameters for i particles */
 462         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 463         vdwioffset0      = 2*nvdwtype*vdwtype[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             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_ps(iq0,jq0);
 510             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,
 512                                          vdwparam+vdwioffset0+vdwjidx0C,
 513                                          vdwparam+vdwioffset0+vdwjidx0D,
 514                                          &c6_00,&c12_00);
 515
 516             /* COULOMB ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq00,rinv00);
 518             felec            = _mm_mul_ps(velec,rinvsq00);
 519
 520             /* LENNARD-JONES DISPERSION/REPULSION */
 521
 522             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 523             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 524
 525             fscal            = _mm_add_ps(felec,fvdw);
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_ps(fscal,dx00);
 529             ty               = _mm_mul_ps(fscal,dy00);
 530             tz               = _mm_mul_ps(fscal,dz00);
 531
 532             /* Update vectorial force */
 533             fix0             = _mm_add_ps(fix0,tx);
 534             fiy0             = _mm_add_ps(fiy0,ty);
 535             fiz0             = _mm_add_ps(fiz0,tz);
 536
 537             fjptrA             = f+j_coord_offsetA;
 538             fjptrB             = f+j_coord_offsetB;
 539             fjptrC             = f+j_coord_offsetC;
 540             fjptrD             = f+j_coord_offsetD;
 541             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 542
 543             /* Inner loop uses 34 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             /* Get j neighbor index, and coordinate index */
 550             jnrlistA         = jjnr[jidx];
 551             jnrlistB         = jjnr[jidx+1];
 552             jnrlistC         = jjnr[jidx+2];
 553             jnrlistD         = jjnr[jidx+3];
 554             /* Sign of each element will be negative for non-real atoms.
 555              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 556              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 557              */
 558             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 559             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 560             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 561             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 562             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 563             j_coord_offsetA  = DIM*jnrA;
 564             j_coord_offsetB  = DIM*jnrB;
 565             j_coord_offsetC  = DIM*jnrC;
 566             j_coord_offsetD  = DIM*jnrD;
 567
 568             /* load j atom coordinates */
 569             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 570                                               x+j_coord_offsetC,x+j_coord_offsetD,
 571                                               &jx0,&jy0,&jz0);
 572
 573             /* Calculate displacement vector */
 574             dx00             = _mm_sub_ps(ix0,jx0);
 575             dy00             = _mm_sub_ps(iy0,jy0);
 576             dz00             = _mm_sub_ps(iz0,jz0);
 577
 578             /* Calculate squared distance and things based on it */
 579             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 580
 581             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 582
 583             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 584
 585             /* Load parameters for j particles */
 586             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 587                                                               charge+jnrC+0,charge+jnrD+0);
 588             vdwjidx0A        = 2*vdwtype[jnrA+0];
 589             vdwjidx0B        = 2*vdwtype[jnrB+0];
 590             vdwjidx0C        = 2*vdwtype[jnrC+0];
 591             vdwjidx0D        = 2*vdwtype[jnrD+0];
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq00             = _mm_mul_ps(iq0,jq0);
 599             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 600                                          vdwparam+vdwioffset0+vdwjidx0B,
 601                                          vdwparam+vdwioffset0+vdwjidx0C,
 602                                          vdwparam+vdwioffset0+vdwjidx0D,
 603                                          &c6_00,&c12_00);
 604
 605             /* COULOMB ELECTROSTATICS */
 606             velec            = _mm_mul_ps(qq00,rinv00);
 607             felec            = _mm_mul_ps(velec,rinvsq00);
 608
 609             /* LENNARD-JONES DISPERSION/REPULSION */
 610
 611             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 612             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 613
 614             fscal            = _mm_add_ps(felec,fvdw);
 615
 616             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm_mul_ps(fscal,dx00);
 620             ty               = _mm_mul_ps(fscal,dy00);
 621             tz               = _mm_mul_ps(fscal,dz00);
 622
 623             /* Update vectorial force */
 624             fix0             = _mm_add_ps(fix0,tx);
 625             fiy0             = _mm_add_ps(fiy0,ty);
 626             fiz0             = _mm_add_ps(fiz0,tz);
 627
 628             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 629             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 630             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 631             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 632             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 633
 634             /* Inner loop uses 34 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 640                                               f+i_coord_offset,fshift+i_shift_offset);
 641
 642         /* Increment number of inner iterations */
 643         inneriter                  += j_index_end - j_index_start;
 644
 645         /* Outer loop uses 7 flops */
 646     }
 647
 648     /* Increment number of outer iterations */
 649     outeriter        += nri;
 650
 651     /* Update outer/inner flops */
 652
 653     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 654 }