src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJEw_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     int              nvdwtype;
  89     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  93     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  94     __m128           c6grid_00;
  95     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  96     real             *vdwgridparam;
  97     __m128           one_half = _mm_set1_ps(0.5);
  98     __m128           minus_one = _mm_set1_ps(-1.0);
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117     vdwgridparam     = fr->ljpme_c6grid;
 118     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 119     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 120     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_ps();
 155         fiy0             = _mm_setzero_ps();
 156         fiz0             = _mm_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         vvdwsum          = _mm_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               x+j_coord_offsetC,x+j_coord_offsetD,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_ps(ix0,jx0);
 185             dy00             = _mm_sub_ps(iy0,jy0);
 186             dz00             = _mm_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 192
 193             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198             vdwjidx0C        = 2*vdwtype[jnrC+0];
 199             vdwjidx0D        = 2*vdwtype[jnrD+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             r00              = _mm_mul_ps(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,
 210                                          vdwparam+vdwioffset0+vdwjidx0C,
 211                                          vdwparam+vdwioffset0+vdwjidx0D,
 212                                          &c6_00,&c12_00);
 213             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 214                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 215                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 216                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 217
 218             /* Analytical LJ-PME */
 219             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 220             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 221             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 222             exponent         = gmx_simd_exp_r(ewcljrsq);
 223             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 224             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 225             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 226             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 227             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 228             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 229             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 230             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);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 234
 235             fscal            = fvdw;
 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             /* Inner loop uses 51 flops */
 254         }
 255
 256         if(jidx<j_index_end)
 257         {
 258
 259             /* Get j neighbor index, and coordinate index */
 260             jnrlistA         = jjnr[jidx];
 261             jnrlistB         = jjnr[jidx+1];
 262             jnrlistC         = jjnr[jidx+2];
 263             jnrlistD         = jjnr[jidx+3];
 264             /* Sign of each element will be negative for non-real atoms.
 265              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 266              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 267              */
 268             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 269             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 270             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 271             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 272             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277
 278             /* load j atom coordinates */
 279             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 280                                               x+j_coord_offsetC,x+j_coord_offsetD,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_ps(ix0,jx0);
 285             dy00             = _mm_sub_ps(iy0,jy0);
 286             dz00             = _mm_sub_ps(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 292
 293             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             vdwjidx0A        = 2*vdwtype[jnrA+0];
 297             vdwjidx0B        = 2*vdwtype[jnrB+0];
 298             vdwjidx0C        = 2*vdwtype[jnrC+0];
 299             vdwjidx0D        = 2*vdwtype[jnrD+0];
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r00              = _mm_mul_ps(rsq00,rinv00);
 306             r00              = _mm_andnot_ps(dummy_mask,r00);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 310                                          vdwparam+vdwioffset0+vdwjidx0B,
 311                                          vdwparam+vdwioffset0+vdwjidx0C,
 312                                          vdwparam+vdwioffset0+vdwjidx0D,
 313                                          &c6_00,&c12_00);
 314             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 315                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 316                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 317                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 318
 319             /* Analytical LJ-PME */
 320             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 321             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 322             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 323             exponent         = gmx_simd_exp_r(ewcljrsq);
 324             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 325             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 326             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 327             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 328             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 329             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 330             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 331             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);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 335             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 336
 337             fscal            = fvdw;
 338
 339             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm_mul_ps(fscal,dx00);
 343             ty               = _mm_mul_ps(fscal,dy00);
 344             tz               = _mm_mul_ps(fscal,dz00);
 345
 346             /* Update vectorial force */
 347             fix0             = _mm_add_ps(fix0,tx);
 348             fiy0             = _mm_add_ps(fiy0,ty);
 349             fiz0             = _mm_add_ps(fiz0,tz);
 350
 351             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 352             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 353             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 354             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 355             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 356
 357             /* Inner loop uses 52 flops */
 358         }
 359
 360         /* End of innermost loop */
 361
 362         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 363                                               f+i_coord_offset,fshift+i_shift_offset);
 364
 365         ggid                        = gid[iidx];
 366         /* Update potential energies */
 367         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 7 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_single
 384  * Electrostatics interaction: None
 385  * VdW interaction:            LJEwald
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_single
 391                     (t_nblist                    * gmx_restrict       nlist,
 392                      rvec                        * gmx_restrict          xx,
 393                      rvec                        * gmx_restrict          ff,
 394                      t_forcerec                  * gmx_restrict          fr,
 395                      t_mdatoms                   * gmx_restrict     mdatoms,
 396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 397                      t_nrnb                      * gmx_restrict        nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB,jnrC,jnrD;
 407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 408     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 409     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 410     real             rcutoff_scalar;
 411     real             *shiftvec,*fshift,*x,*f;
 412     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 413     real             scratch[4*DIM];
 414     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 415     int              vdwioffset0;
 416     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 417     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 418     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 419     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 420     int              nvdwtype;
 421     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 422     int              *vdwtype;
 423     real             *vdwparam;
 424     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 425     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 426     __m128           c6grid_00;
 427     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 428     real             *vdwgridparam;
 429     __m128           one_half = _mm_set1_ps(0.5);
 430     __m128           minus_one = _mm_set1_ps(-1.0);
 431     __m128           dummy_mask,cutoff_mask;
 432     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 433     __m128           one     = _mm_set1_ps(1.0);
 434     __m128           two     = _mm_set1_ps(2.0);
 435     x                = xx[0];
 436     f                = ff[0];
 437
 438     nri              = nlist->nri;
 439     iinr             = nlist->iinr;
 440     jindex           = nlist->jindex;
 441     jjnr             = nlist->jjnr;
 442     shiftidx         = nlist->shift;
 443     gid              = nlist->gid;
 444     shiftvec         = fr->shift_vec[0];
 445     fshift           = fr->fshift[0];
 446     nvdwtype         = fr->ntype;
 447     vdwparam         = fr->nbfp;
 448     vdwtype          = mdatoms->typeA;
 449     vdwgridparam     = fr->ljpme_c6grid;
 450     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 451     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 452     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 453
 454     /* Avoid stupid compiler warnings */
 455     jnrA = jnrB = jnrC = jnrD = 0;
 456     j_coord_offsetA = 0;
 457     j_coord_offsetB = 0;
 458     j_coord_offsetC = 0;
 459     j_coord_offsetD = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     for(iidx=0;iidx<4*DIM;iidx++)
 465     {
 466         scratch[iidx] = 0.0;
 467     }
 468
 469     /* Start outer loop over neighborlists */
 470     for(iidx=0; iidx<nri; iidx++)
 471     {
 472         /* Load shift vector for this list */
 473         i_shift_offset   = DIM*shiftidx[iidx];
 474
 475         /* Load limits for loop over neighbors */
 476         j_index_start    = jindex[iidx];
 477         j_index_end      = jindex[iidx+1];
 478
 479         /* Get outer coordinate index */
 480         inr              = iinr[iidx];
 481         i_coord_offset   = DIM*inr;
 482
 483         /* Load i particle coords and add shift vector */
 484         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 485
 486         fix0             = _mm_setzero_ps();
 487         fiy0             = _mm_setzero_ps();
 488         fiz0             = _mm_setzero_ps();
 489
 490         /* Load parameters for i particles */
 491         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             jnrC             = jjnr[jidx+2];
 501             jnrD             = jjnr[jidx+3];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504             j_coord_offsetC  = DIM*jnrC;
 505             j_coord_offsetD  = DIM*jnrD;
 506
 507             /* load j atom coordinates */
 508             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                               x+j_coord_offsetC,x+j_coord_offsetD,
 510                                               &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _mm_sub_ps(ix0,jx0);
 514             dy00             = _mm_sub_ps(iy0,jy0);
 515             dz00             = _mm_sub_ps(iz0,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 519
 520             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 521
 522             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 523
 524             /* Load parameters for j particles */
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527             vdwjidx0C        = 2*vdwtype[jnrC+0];
 528             vdwjidx0D        = 2*vdwtype[jnrD+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm_mul_ps(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 538                                          vdwparam+vdwioffset0+vdwjidx0B,
 539                                          vdwparam+vdwioffset0+vdwjidx0C,
 540                                          vdwparam+vdwioffset0+vdwjidx0D,
 541                                          &c6_00,&c12_00);
 542             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 546
 547             /* Analytical LJ-PME */
 548             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 549             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 550             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 551             exponent         = gmx_simd_exp_r(ewcljrsq);
 552             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 553             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 554             /* f6A = 6 * C6grid * (1 - poly) */
 555             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 556             /* f6B = C6grid * exponent * beta^6 */
 557             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 558             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 559             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);
 560
 561             fscal            = fvdw;
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm_mul_ps(fscal,dx00);
 565             ty               = _mm_mul_ps(fscal,dy00);
 566             tz               = _mm_mul_ps(fscal,dz00);
 567
 568             /* Update vectorial force */
 569             fix0             = _mm_add_ps(fix0,tx);
 570             fiy0             = _mm_add_ps(fiy0,ty);
 571             fiz0             = _mm_add_ps(fiz0,tz);
 572
 573             fjptrA             = f+j_coord_offsetA;
 574             fjptrB             = f+j_coord_offsetB;
 575             fjptrC             = f+j_coord_offsetC;
 576             fjptrD             = f+j_coord_offsetD;
 577             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 578
 579             /* Inner loop uses 46 flops */
 580         }
 581
 582         if(jidx<j_index_end)
 583         {
 584
 585             /* Get j neighbor index, and coordinate index */
 586             jnrlistA         = jjnr[jidx];
 587             jnrlistB         = jjnr[jidx+1];
 588             jnrlistC         = jjnr[jidx+2];
 589             jnrlistD         = jjnr[jidx+3];
 590             /* Sign of each element will be negative for non-real atoms.
 591              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 592              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 593              */
 594             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 595             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 596             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 597             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 598             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 599             j_coord_offsetA  = DIM*jnrA;
 600             j_coord_offsetB  = DIM*jnrB;
 601             j_coord_offsetC  = DIM*jnrC;
 602             j_coord_offsetD  = DIM*jnrD;
 603
 604             /* load j atom coordinates */
 605             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 606                                               x+j_coord_offsetC,x+j_coord_offsetD,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _mm_sub_ps(ix0,jx0);
 611             dy00             = _mm_sub_ps(iy0,jy0);
 612             dz00             = _mm_sub_ps(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 616
 617             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 618
 619             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 620
 621             /* Load parameters for j particles */
 622             vdwjidx0A        = 2*vdwtype[jnrA+0];
 623             vdwjidx0B        = 2*vdwtype[jnrB+0];
 624             vdwjidx0C        = 2*vdwtype[jnrC+0];
 625             vdwjidx0D        = 2*vdwtype[jnrD+0];
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r00              = _mm_mul_ps(rsq00,rinv00);
 632             r00              = _mm_andnot_ps(dummy_mask,r00);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 636                                          vdwparam+vdwioffset0+vdwjidx0B,
 637                                          vdwparam+vdwioffset0+vdwjidx0C,
 638                                          vdwparam+vdwioffset0+vdwjidx0D,
 639                                          &c6_00,&c12_00);
 640             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 641                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 642                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 643                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 644
 645             /* Analytical LJ-PME */
 646             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 647             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 648             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 649             exponent         = gmx_simd_exp_r(ewcljrsq);
 650             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 651             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 652             /* f6A = 6 * C6grid * (1 - poly) */
 653             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 654             /* f6B = C6grid * exponent * beta^6 */
 655             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 656             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 657             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);
 658
 659             fscal            = fvdw;
 660
 661             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm_mul_ps(fscal,dx00);
 665             ty               = _mm_mul_ps(fscal,dy00);
 666             tz               = _mm_mul_ps(fscal,dz00);
 667
 668             /* Update vectorial force */
 669             fix0             = _mm_add_ps(fix0,tx);
 670             fiy0             = _mm_add_ps(fiy0,ty);
 671             fiz0             = _mm_add_ps(fiz0,tz);
 672
 673             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 674             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 675             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 676             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 677             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 678
 679             /* Inner loop uses 47 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 6 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 699 }