src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 107     rcutoff_scalar   = fr->rcoulomb;
 108     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 109     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 110
 111     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 112     rvdw             = _mm_set1_ps(fr->rvdw);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_ps();
 147         fiy0             = _mm_setzero_ps();
 148         fiz0             = _mm_setzero_ps();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156         vvdwsum          = _mm_setzero_ps();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             jnrC             = jjnr[jidx+2];
 166             jnrD             = jjnr[jidx+3];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               x+j_coord_offsetC,x+j_coord_offsetD,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_ps(ix0,jx0);
 179             dy00             = _mm_sub_ps(iy0,jy0);
 180             dz00             = _mm_sub_ps(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 186
 187             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 191                                                               charge+jnrC+0,charge+jnrD+0);
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194             vdwjidx0C        = 2*vdwtype[jnrC+0];
 195             vdwjidx0D        = 2*vdwtype[jnrD+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_ps(iq0,jq0);
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* REACTION-FIELD ELECTROSTATICS */
 213             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 214             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 215
 216             /* LENNARD-JONES DISPERSION/REPULSION */
 217
 218             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 219             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 220             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 221             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 222                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 223             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 224
 225             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velec            = _mm_and_ps(velec,cutoff_mask);
 229             velecsum         = _mm_add_ps(velecsum,velec);
 230             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 231             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 232
 233             fscal            = _mm_add_ps(felec,fvdw);
 234
 235             fscal            = _mm_and_ps(fscal,cutoff_mask);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm_mul_ps(fscal,dx00);
 239             ty               = _mm_mul_ps(fscal,dy00);
 240             tz               = _mm_mul_ps(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_add_ps(fix0,tx);
 244             fiy0             = _mm_add_ps(fiy0,ty);
 245             fiz0             = _mm_add_ps(fiz0,tz);
 246
 247             fjptrA             = f+j_coord_offsetA;
 248             fjptrB             = f+j_coord_offsetB;
 249             fjptrC             = f+j_coord_offsetC;
 250             fjptrD             = f+j_coord_offsetD;
 251             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 252
 253             }
 254
 255             /* Inner loop uses 54 flops */
 256         }
 257
 258         if(jidx<j_index_end)
 259         {
 260
 261             /* Get j neighbor index, and coordinate index */
 262             jnrlistA         = jjnr[jidx];
 263             jnrlistB         = jjnr[jidx+1];
 264             jnrlistC         = jjnr[jidx+2];
 265             jnrlistD         = jjnr[jidx+3];
 266             /* Sign of each element will be negative for non-real atoms.
 267              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 268              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 269              */
 270             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 271             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 272             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 273             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 274             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 275             j_coord_offsetA  = DIM*jnrA;
 276             j_coord_offsetB  = DIM*jnrB;
 277             j_coord_offsetC  = DIM*jnrC;
 278             j_coord_offsetD  = DIM*jnrD;
 279
 280             /* load j atom coordinates */
 281             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 282                                               x+j_coord_offsetC,x+j_coord_offsetD,
 283                                               &jx0,&jy0,&jz0);
 284
 285             /* Calculate displacement vector */
 286             dx00             = _mm_sub_ps(ix0,jx0);
 287             dy00             = _mm_sub_ps(iy0,jy0);
 288             dz00             = _mm_sub_ps(iz0,jz0);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 292
 293             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 294
 295             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 296
 297             /* Load parameters for j particles */
 298             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 299                                                               charge+jnrC+0,charge+jnrD+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301             vdwjidx0B        = 2*vdwtype[jnrB+0];
 302             vdwjidx0C        = 2*vdwtype[jnrC+0];
 303             vdwjidx0D        = 2*vdwtype[jnrD+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq00,rcutoff2))
 310             {
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq00             = _mm_mul_ps(iq0,jq0);
 314             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 315                                          vdwparam+vdwioffset0+vdwjidx0B,
 316                                          vdwparam+vdwioffset0+vdwjidx0C,
 317                                          vdwparam+vdwioffset0+vdwjidx0D,
 318                                          &c6_00,&c12_00);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 322             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 328             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 329             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 330                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 331             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 332
 333             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_ps(velec,cutoff_mask);
 337             velec            = _mm_andnot_ps(dummy_mask,velec);
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 340             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 341             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 342
 343             fscal            = _mm_add_ps(felec,fvdw);
 344
 345             fscal            = _mm_and_ps(fscal,cutoff_mask);
 346
 347             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx00);
 351             ty               = _mm_mul_ps(fscal,dy00);
 352             tz               = _mm_mul_ps(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm_add_ps(fix0,tx);
 356             fiy0             = _mm_add_ps(fiy0,ty);
 357             fiz0             = _mm_add_ps(fiz0,tz);
 358
 359             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 360             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 361             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 362             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 364
 365             }
 366
 367             /* Inner loop uses 54 flops */
 368         }
 369
 370         /* End of innermost loop */
 371
 372         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 373                                               f+i_coord_offset,fshift+i_shift_offset);
 374
 375         ggid                        = gid[iidx];
 376         /* Update potential energies */
 377         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 378         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 379
 380         /* Increment number of inner iterations */
 381         inneriter                  += j_index_end - j_index_start;
 382
 383         /* Outer loop uses 9 flops */
 384     }
 385
 386     /* Increment number of outer iterations */
 387     outeriter        += nri;
 388
 389     /* Update outer/inner flops */
 390
 391     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 392 }
 393 /*
 394  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_single
 395  * Electrostatics interaction: ReactionField
 396  * VdW interaction:            LennardJones
 397  * Geometry:                   Particle-Particle
 398  * Calculate force/pot:        Force
 399  */
 400 void
 401 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_single
 402                     (t_nblist * gmx_restrict                nlist,
 403                      rvec * gmx_restrict                    xx,
 404                      rvec * gmx_restrict                    ff,
 405                      t_forcerec * gmx_restrict              fr,
 406                      t_mdatoms * gmx_restrict               mdatoms,
 407                      nb_kernel_data_t * gmx_restrict        kernel_data,
 408                      t_nrnb * gmx_restrict                  nrnb)
 409 {
 410     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 411      * just 0 for non-waters.
 412      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 413      * jnr indices corresponding to data put in the four positions in the SIMD register.
 414      */
 415     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 416     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 417     int              jnrA,jnrB,jnrC,jnrD;
 418     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 419     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 420     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 421     real             rcutoff_scalar;
 422     real             *shiftvec,*fshift,*x,*f;
 423     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 424     real             scratch[4*DIM];
 425     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 426     int              vdwioffset0;
 427     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 428     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 429     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 430     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 431     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 432     real             *charge;
 433     int              nvdwtype;
 434     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 435     int              *vdwtype;
 436     real             *vdwparam;
 437     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 438     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 439     __m128           dummy_mask,cutoff_mask;
 440     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 441     __m128           one     = _mm_set1_ps(1.0);
 442     __m128           two     = _mm_set1_ps(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm_set1_ps(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456     krf              = _mm_set1_ps(fr->ic->k_rf);
 457     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 458     crf              = _mm_set1_ps(fr->ic->c_rf);
 459     nvdwtype         = fr->ntype;
 460     vdwparam         = fr->nbfp;
 461     vdwtype          = mdatoms->typeA;
 462
 463     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 464     rcutoff_scalar   = fr->rcoulomb;
 465     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 466     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 467
 468     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 469     rvdw             = _mm_set1_ps(fr->rvdw);
 470
 471     /* Avoid stupid compiler warnings */
 472     jnrA = jnrB = jnrC = jnrD = 0;
 473     j_coord_offsetA = 0;
 474     j_coord_offsetB = 0;
 475     j_coord_offsetC = 0;
 476     j_coord_offsetD = 0;
 477
 478     outeriter        = 0;
 479     inneriter        = 0;
 480
 481     for(iidx=0;iidx<4*DIM;iidx++)
 482     {
 483         scratch[iidx] = 0.0;
 484     }
 485
 486     /* Start outer loop over neighborlists */
 487     for(iidx=0; iidx<nri; iidx++)
 488     {
 489         /* Load shift vector for this list */
 490         i_shift_offset   = DIM*shiftidx[iidx];
 491
 492         /* Load limits for loop over neighbors */
 493         j_index_start    = jindex[iidx];
 494         j_index_end      = jindex[iidx+1];
 495
 496         /* Get outer coordinate index */
 497         inr              = iinr[iidx];
 498         i_coord_offset   = DIM*inr;
 499
 500         /* Load i particle coords and add shift vector */
 501         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 502
 503         fix0             = _mm_setzero_ps();
 504         fiy0             = _mm_setzero_ps();
 505         fiz0             = _mm_setzero_ps();
 506
 507         /* Load parameters for i particles */
 508         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 509         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 510
 511         /* Start inner kernel loop */
 512         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 513         {
 514
 515             /* Get j neighbor index, and coordinate index */
 516             jnrA             = jjnr[jidx];
 517             jnrB             = jjnr[jidx+1];
 518             jnrC             = jjnr[jidx+2];
 519             jnrD             = jjnr[jidx+3];
 520             j_coord_offsetA  = DIM*jnrA;
 521             j_coord_offsetB  = DIM*jnrB;
 522             j_coord_offsetC  = DIM*jnrC;
 523             j_coord_offsetD  = DIM*jnrD;
 524
 525             /* load j atom coordinates */
 526             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 527                                               x+j_coord_offsetC,x+j_coord_offsetD,
 528                                               &jx0,&jy0,&jz0);
 529
 530             /* Calculate displacement vector */
 531             dx00             = _mm_sub_ps(ix0,jx0);
 532             dy00             = _mm_sub_ps(iy0,jy0);
 533             dz00             = _mm_sub_ps(iz0,jz0);
 534
 535             /* Calculate squared distance and things based on it */
 536             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 537
 538             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 539
 540             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 541
 542             /* Load parameters for j particles */
 543             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 544                                                               charge+jnrC+0,charge+jnrD+0);
 545             vdwjidx0A        = 2*vdwtype[jnrA+0];
 546             vdwjidx0B        = 2*vdwtype[jnrB+0];
 547             vdwjidx0C        = 2*vdwtype[jnrC+0];
 548             vdwjidx0D        = 2*vdwtype[jnrD+0];
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             if (gmx_mm_any_lt(rsq00,rcutoff2))
 555             {
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_ps(iq0,jq0);
 559             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,
 561                                          vdwparam+vdwioffset0+vdwjidx0C,
 562                                          vdwparam+vdwioffset0+vdwjidx0D,
 563                                          &c6_00,&c12_00);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 572
 573             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 574
 575             fscal            = _mm_add_ps(felec,fvdw);
 576
 577             fscal            = _mm_and_ps(fscal,cutoff_mask);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx00);
 581             ty               = _mm_mul_ps(fscal,dy00);
 582             tz               = _mm_mul_ps(fscal,dz00);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_add_ps(fix0,tx);
 586             fiy0             = _mm_add_ps(fiy0,ty);
 587             fiz0             = _mm_add_ps(fiz0,tz);
 588
 589             fjptrA             = f+j_coord_offsetA;
 590             fjptrB             = f+j_coord_offsetB;
 591             fjptrC             = f+j_coord_offsetC;
 592             fjptrD             = f+j_coord_offsetD;
 593             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 594
 595             }
 596
 597             /* Inner loop uses 37 flops */
 598         }
 599
 600         if(jidx<j_index_end)
 601         {
 602
 603             /* Get j neighbor index, and coordinate index */
 604             jnrlistA         = jjnr[jidx];
 605             jnrlistB         = jjnr[jidx+1];
 606             jnrlistC         = jjnr[jidx+2];
 607             jnrlistD         = jjnr[jidx+3];
 608             /* Sign of each element will be negative for non-real atoms.
 609              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 610              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 611              */
 612             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 613             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 614             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 615             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 616             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 617             j_coord_offsetA  = DIM*jnrA;
 618             j_coord_offsetB  = DIM*jnrB;
 619             j_coord_offsetC  = DIM*jnrC;
 620             j_coord_offsetD  = DIM*jnrD;
 621
 622             /* load j atom coordinates */
 623             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                               x+j_coord_offsetC,x+j_coord_offsetD,
 625                                               &jx0,&jy0,&jz0);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm_sub_ps(ix0,jx0);
 629             dy00             = _mm_sub_ps(iy0,jy0);
 630             dz00             = _mm_sub_ps(iz0,jz0);
 631
 632             /* Calculate squared distance and things based on it */
 633             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 634
 635             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 636
 637             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 638
 639             /* Load parameters for j particles */
 640             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 641                                                               charge+jnrC+0,charge+jnrD+0);
 642             vdwjidx0A        = 2*vdwtype[jnrA+0];
 643             vdwjidx0B        = 2*vdwtype[jnrB+0];
 644             vdwjidx0C        = 2*vdwtype[jnrC+0];
 645             vdwjidx0D        = 2*vdwtype[jnrD+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm_any_lt(rsq00,rcutoff2))
 652             {
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq00             = _mm_mul_ps(iq0,jq0);
 656             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 657                                          vdwparam+vdwioffset0+vdwjidx0B,
 658                                          vdwparam+vdwioffset0+vdwjidx0C,
 659                                          vdwparam+vdwioffset0+vdwjidx0D,
 660                                          &c6_00,&c12_00);
 661
 662             /* REACTION-FIELD ELECTROSTATICS */
 663             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 664
 665             /* LENNARD-JONES DISPERSION/REPULSION */
 666
 667             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 668             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 669
 670             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 671
 672             fscal            = _mm_add_ps(felec,fvdw);
 673
 674             fscal            = _mm_and_ps(fscal,cutoff_mask);
 675
 676             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm_mul_ps(fscal,dx00);
 680             ty               = _mm_mul_ps(fscal,dy00);
 681             tz               = _mm_mul_ps(fscal,dz00);
 682
 683             /* Update vectorial force */
 684             fix0             = _mm_add_ps(fix0,tx);
 685             fiy0             = _mm_add_ps(fiy0,ty);
 686             fiz0             = _mm_add_ps(fiz0,tz);
 687
 688             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 689             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 690             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 691             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 692             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 693
 694             }
 695
 696             /* Inner loop uses 37 flops */
 697         }
 698
 699         /* End of innermost loop */
 700
 701         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 702                                               f+i_coord_offset,fshift+i_shift_offset);
 703
 704         /* Increment number of inner iterations */
 705         inneriter                  += j_index_end - j_index_start;
 706
 707         /* Outer loop uses 7 flops */
 708     }
 709
 710     /* Increment number of outer iterations */
 711     outeriter        += nri;
 712
 713     /* Update outer/inner flops */
 714
 715     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 716 }