src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  92     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  93     __m128           c6grid_00;
  94     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     real             *vdwgridparam;
  96     __m128           one_half = _mm_set1_ps(0.5);
  97     __m128           minus_one = _mm_set1_ps(-1.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116     vdwgridparam     = fr->ljpme_c6grid;
 117     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 118     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 119     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 120
 121     rcutoff_scalar   = fr->ic->rvdw;
 122     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 123     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 124
 125     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 126     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163
 164         /* Load parameters for i particles */
 165         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 166
 167         /* Reset potential sums */
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 196
 197             rinv00           = sse2_invsqrt_f(rsq00);
 198
 199             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             vdwjidx0A        = 2*vdwtype[jnrA+0];
 203             vdwjidx0B        = 2*vdwtype[jnrB+0];
 204             vdwjidx0C        = 2*vdwtype[jnrC+0];
 205             vdwjidx0D        = 2*vdwtype[jnrD+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             if (gmx_mm_any_lt(rsq00,rcutoff2))
 212             {
 213
 214             r00              = _mm_mul_ps(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,
 219                                          vdwparam+vdwioffset0+vdwjidx0C,
 220                                          vdwparam+vdwioffset0+vdwjidx0D,
 221                                          &c6_00,&c12_00);
 222             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 223                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 224                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 225                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 226
 227             /* Analytical LJ-PME */
 228             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 229             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 230             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 231             exponent         = sse2_exp_f(ewcljrsq);
 232             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 233             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 234             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 235             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 236             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 237             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) ,
 238                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 239             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 240             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 241
 242             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 246             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 247
 248             fscal            = fvdw;
 249
 250             fscal            = _mm_and_ps(fscal,cutoff_mask);
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_ps(fscal,dx00);
 254             ty               = _mm_mul_ps(fscal,dy00);
 255             tz               = _mm_mul_ps(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_ps(fix0,tx);
 259             fiy0             = _mm_add_ps(fiy0,ty);
 260             fiz0             = _mm_add_ps(fiz0,tz);
 261
 262             fjptrA             = f+j_coord_offsetA;
 263             fjptrB             = f+j_coord_offsetB;
 264             fjptrC             = f+j_coord_offsetC;
 265             fjptrD             = f+j_coord_offsetD;
 266             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 267
 268             }
 269
 270             /* Inner loop uses 62 flops */
 271         }
 272
 273         if(jidx<j_index_end)
 274         {
 275
 276             /* Get j neighbor index, and coordinate index */
 277             jnrlistA         = jjnr[jidx];
 278             jnrlistB         = jjnr[jidx+1];
 279             jnrlistC         = jjnr[jidx+2];
 280             jnrlistD         = jjnr[jidx+3];
 281             /* Sign of each element will be negative for non-real atoms.
 282              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 283              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 284              */
 285             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 286             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 287             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 288             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 289             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 290             j_coord_offsetA  = DIM*jnrA;
 291             j_coord_offsetB  = DIM*jnrB;
 292             j_coord_offsetC  = DIM*jnrC;
 293             j_coord_offsetD  = DIM*jnrD;
 294
 295             /* load j atom coordinates */
 296             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 297                                               x+j_coord_offsetC,x+j_coord_offsetD,
 298                                               &jx0,&jy0,&jz0);
 299
 300             /* Calculate displacement vector */
 301             dx00             = _mm_sub_ps(ix0,jx0);
 302             dy00             = _mm_sub_ps(iy0,jy0);
 303             dz00             = _mm_sub_ps(iz0,jz0);
 304
 305             /* Calculate squared distance and things based on it */
 306             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 307
 308             rinv00           = sse2_invsqrt_f(rsq00);
 309
 310             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 311
 312             /* Load parameters for j particles */
 313             vdwjidx0A        = 2*vdwtype[jnrA+0];
 314             vdwjidx0B        = 2*vdwtype[jnrB+0];
 315             vdwjidx0C        = 2*vdwtype[jnrC+0];
 316             vdwjidx0D        = 2*vdwtype[jnrD+0];
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             if (gmx_mm_any_lt(rsq00,rcutoff2))
 323             {
 324
 325             r00              = _mm_mul_ps(rsq00,rinv00);
 326             r00              = _mm_andnot_ps(dummy_mask,r00);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 330                                          vdwparam+vdwioffset0+vdwjidx0B,
 331                                          vdwparam+vdwioffset0+vdwjidx0C,
 332                                          vdwparam+vdwioffset0+vdwjidx0D,
 333                                          &c6_00,&c12_00);
 334             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 335                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 336                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 337                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 338
 339             /* Analytical LJ-PME */
 340             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 341             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 342             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 343             exponent         = sse2_exp_f(ewcljrsq);
 344             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 345             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 346             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 347             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 348             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 349             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) ,
 350                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 351             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 352             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 353
 354             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 358             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 359             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 360
 361             fscal            = fvdw;
 362
 363             fscal            = _mm_and_ps(fscal,cutoff_mask);
 364
 365             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm_mul_ps(fscal,dx00);
 369             ty               = _mm_mul_ps(fscal,dy00);
 370             tz               = _mm_mul_ps(fscal,dz00);
 371
 372             /* Update vectorial force */
 373             fix0             = _mm_add_ps(fix0,tx);
 374             fiy0             = _mm_add_ps(fiy0,ty);
 375             fiz0             = _mm_add_ps(fiz0,tz);
 376
 377             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 378             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 379             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 380             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 381             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 382
 383             }
 384
 385             /* Inner loop uses 63 flops */
 386         }
 387
 388         /* End of innermost loop */
 389
 390         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 391                                               f+i_coord_offset,fshift+i_shift_offset);
 392
 393         ggid                        = gid[iidx];
 394         /* Update potential energies */
 395         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 396
 397         /* Increment number of inner iterations */
 398         inneriter                  += j_index_end - j_index_start;
 399
 400         /* Outer loop uses 7 flops */
 401     }
 402
 403     /* Increment number of outer iterations */
 404     outeriter        += nri;
 405
 406     /* Update outer/inner flops */
 407
 408     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*63);
 409 }
 410 /*
 411  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_single
 412  * Electrostatics interaction: None
 413  * VdW interaction:            LJEwald
 414  * Geometry:                   Particle-Particle
 415  * Calculate force/pot:        Force
 416  */
 417 void
 418 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_single
 419                     (t_nblist                    * gmx_restrict       nlist,
 420                      rvec                        * gmx_restrict          xx,
 421                      rvec                        * gmx_restrict          ff,
 422                      struct t_forcerec           * gmx_restrict          fr,
 423                      t_mdatoms                   * gmx_restrict     mdatoms,
 424                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 425                      t_nrnb                      * gmx_restrict        nrnb)
 426 {
 427     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 428      * just 0 for non-waters.
 429      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 430      * jnr indices corresponding to data put in the four positions in the SIMD register.
 431      */
 432     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 433     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 434     int              jnrA,jnrB,jnrC,jnrD;
 435     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 436     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 437     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 438     real             rcutoff_scalar;
 439     real             *shiftvec,*fshift,*x,*f;
 440     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 441     real             scratch[4*DIM];
 442     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 443     int              vdwioffset0;
 444     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 445     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 446     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 447     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 448     int              nvdwtype;
 449     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 450     int              *vdwtype;
 451     real             *vdwparam;
 452     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 453     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 454     __m128           c6grid_00;
 455     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 456     real             *vdwgridparam;
 457     __m128           one_half = _mm_set1_ps(0.5);
 458     __m128           minus_one = _mm_set1_ps(-1.0);
 459     __m128           dummy_mask,cutoff_mask;
 460     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 461     __m128           one     = _mm_set1_ps(1.0);
 462     __m128           two     = _mm_set1_ps(2.0);
 463     x                = xx[0];
 464     f                = ff[0];
 465
 466     nri              = nlist->nri;
 467     iinr             = nlist->iinr;
 468     jindex           = nlist->jindex;
 469     jjnr             = nlist->jjnr;
 470     shiftidx         = nlist->shift;
 471     gid              = nlist->gid;
 472     shiftvec         = fr->shift_vec[0];
 473     fshift           = fr->fshift[0];
 474     nvdwtype         = fr->ntype;
 475     vdwparam         = fr->nbfp;
 476     vdwtype          = mdatoms->typeA;
 477     vdwgridparam     = fr->ljpme_c6grid;
 478     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 479     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 480     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 481
 482     rcutoff_scalar   = fr->ic->rvdw;
 483     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 484     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 485
 486     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 487     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 488
 489     /* Avoid stupid compiler warnings */
 490     jnrA = jnrB = jnrC = jnrD = 0;
 491     j_coord_offsetA = 0;
 492     j_coord_offsetB = 0;
 493     j_coord_offsetC = 0;
 494     j_coord_offsetD = 0;
 495
 496     outeriter        = 0;
 497     inneriter        = 0;
 498
 499     for(iidx=0;iidx<4*DIM;iidx++)
 500     {
 501         scratch[iidx] = 0.0;
 502     }
 503
 504     /* Start outer loop over neighborlists */
 505     for(iidx=0; iidx<nri; iidx++)
 506     {
 507         /* Load shift vector for this list */
 508         i_shift_offset   = DIM*shiftidx[iidx];
 509
 510         /* Load limits for loop over neighbors */
 511         j_index_start    = jindex[iidx];
 512         j_index_end      = jindex[iidx+1];
 513
 514         /* Get outer coordinate index */
 515         inr              = iinr[iidx];
 516         i_coord_offset   = DIM*inr;
 517
 518         /* Load i particle coords and add shift vector */
 519         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 520
 521         fix0             = _mm_setzero_ps();
 522         fiy0             = _mm_setzero_ps();
 523         fiz0             = _mm_setzero_ps();
 524
 525         /* Load parameters for i particles */
 526         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 527
 528         /* Start inner kernel loop */
 529         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 530         {
 531
 532             /* Get j neighbor index, and coordinate index */
 533             jnrA             = jjnr[jidx];
 534             jnrB             = jjnr[jidx+1];
 535             jnrC             = jjnr[jidx+2];
 536             jnrD             = jjnr[jidx+3];
 537             j_coord_offsetA  = DIM*jnrA;
 538             j_coord_offsetB  = DIM*jnrB;
 539             j_coord_offsetC  = DIM*jnrC;
 540             j_coord_offsetD  = DIM*jnrD;
 541
 542             /* load j atom coordinates */
 543             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 544                                               x+j_coord_offsetC,x+j_coord_offsetD,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_ps(ix0,jx0);
 549             dy00             = _mm_sub_ps(iy0,jy0);
 550             dz00             = _mm_sub_ps(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 554
 555             rinv00           = sse2_invsqrt_f(rsq00);
 556
 557             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             vdwjidx0A        = 2*vdwtype[jnrA+0];
 561             vdwjidx0B        = 2*vdwtype[jnrB+0];
 562             vdwjidx0C        = 2*vdwtype[jnrC+0];
 563             vdwjidx0D        = 2*vdwtype[jnrD+0];
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq00,rcutoff2))
 570             {
 571
 572             r00              = _mm_mul_ps(rsq00,rinv00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 576                                          vdwparam+vdwioffset0+vdwjidx0B,
 577                                          vdwparam+vdwioffset0+vdwjidx0C,
 578                                          vdwparam+vdwioffset0+vdwjidx0D,
 579                                          &c6_00,&c12_00);
 580             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 581                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 582                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 583                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 584
 585             /* Analytical LJ-PME */
 586             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 587             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 588             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 589             exponent         = sse2_exp_f(ewcljrsq);
 590             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 591             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 592             /* f6A = 6 * C6grid * (1 - poly) */
 593             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 594             /* f6B = C6grid * exponent * beta^6 */
 595             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 596             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 597             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 598
 599             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 600
 601             fscal            = fvdw;
 602
 603             fscal            = _mm_and_ps(fscal,cutoff_mask);
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm_mul_ps(fscal,dx00);
 607             ty               = _mm_mul_ps(fscal,dy00);
 608             tz               = _mm_mul_ps(fscal,dz00);
 609
 610             /* Update vectorial force */
 611             fix0             = _mm_add_ps(fix0,tx);
 612             fiy0             = _mm_add_ps(fiy0,ty);
 613             fiz0             = _mm_add_ps(fiz0,tz);
 614
 615             fjptrA             = f+j_coord_offsetA;
 616             fjptrB             = f+j_coord_offsetB;
 617             fjptrC             = f+j_coord_offsetC;
 618             fjptrD             = f+j_coord_offsetD;
 619             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 620
 621             }
 622
 623             /* Inner loop uses 49 flops */
 624         }
 625
 626         if(jidx<j_index_end)
 627         {
 628
 629             /* Get j neighbor index, and coordinate index */
 630             jnrlistA         = jjnr[jidx];
 631             jnrlistB         = jjnr[jidx+1];
 632             jnrlistC         = jjnr[jidx+2];
 633             jnrlistD         = jjnr[jidx+3];
 634             /* Sign of each element will be negative for non-real atoms.
 635              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 636              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 637              */
 638             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 639             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 640             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 641             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 642             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 643             j_coord_offsetA  = DIM*jnrA;
 644             j_coord_offsetB  = DIM*jnrB;
 645             j_coord_offsetC  = DIM*jnrC;
 646             j_coord_offsetD  = DIM*jnrD;
 647
 648             /* load j atom coordinates */
 649             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 650                                               x+j_coord_offsetC,x+j_coord_offsetD,
 651                                               &jx0,&jy0,&jz0);
 652
 653             /* Calculate displacement vector */
 654             dx00             = _mm_sub_ps(ix0,jx0);
 655             dy00             = _mm_sub_ps(iy0,jy0);
 656             dz00             = _mm_sub_ps(iz0,jz0);
 657
 658             /* Calculate squared distance and things based on it */
 659             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 660
 661             rinv00           = sse2_invsqrt_f(rsq00);
 662
 663             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 664
 665             /* Load parameters for j particles */
 666             vdwjidx0A        = 2*vdwtype[jnrA+0];
 667             vdwjidx0B        = 2*vdwtype[jnrB+0];
 668             vdwjidx0C        = 2*vdwtype[jnrC+0];
 669             vdwjidx0D        = 2*vdwtype[jnrD+0];
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             if (gmx_mm_any_lt(rsq00,rcutoff2))
 676             {
 677
 678             r00              = _mm_mul_ps(rsq00,rinv00);
 679             r00              = _mm_andnot_ps(dummy_mask,r00);
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 683                                          vdwparam+vdwioffset0+vdwjidx0B,
 684                                          vdwparam+vdwioffset0+vdwjidx0C,
 685                                          vdwparam+vdwioffset0+vdwjidx0D,
 686                                          &c6_00,&c12_00);
 687             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 688                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 689                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 690                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 691
 692             /* Analytical LJ-PME */
 693             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 694             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 695             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 696             exponent         = sse2_exp_f(ewcljrsq);
 697             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 698             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 699             /* f6A = 6 * C6grid * (1 - poly) */
 700             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 701             /* f6B = C6grid * exponent * beta^6 */
 702             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 703             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 704             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 705
 706             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 707
 708             fscal            = fvdw;
 709
 710             fscal            = _mm_and_ps(fscal,cutoff_mask);
 711
 712             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm_mul_ps(fscal,dx00);
 716             ty               = _mm_mul_ps(fscal,dy00);
 717             tz               = _mm_mul_ps(fscal,dz00);
 718
 719             /* Update vectorial force */
 720             fix0             = _mm_add_ps(fix0,tx);
 721             fiy0             = _mm_add_ps(fiy0,ty);
 722             fiz0             = _mm_add_ps(fiz0,tz);
 723
 724             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 725             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 726             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 727             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 728             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 729
 730             }
 731
 732             /* Inner loop uses 50 flops */
 733         }
 734
 735         /* End of innermost loop */
 736
 737         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 738                                               f+i_coord_offset,fshift+i_shift_offset);
 739
 740         /* Increment number of inner iterations */
 741         inneriter                  += j_index_end - j_index_start;
 742
 743         /* Outer loop uses 6 flops */
 744     }
 745
 746     /* Increment number of outer iterations */
 747     outeriter        += nri;
 748
 749     /* Update outer/inner flops */
 750
 751     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);
 752 }