src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwLJEw_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_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_sse4_1_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
 214             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 215                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 216                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 217                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 218
 219             /* Analytical LJ-PME */
 220             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 221             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 222             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 223             exponent         = gmx_simd_exp_r(ewcljrsq);
 224             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 225             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 226             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 227             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 228             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 229             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 230             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 231             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);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 235
 236             fscal            = fvdw;
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm_mul_ps(fscal,dx00);
 240             ty               = _mm_mul_ps(fscal,dy00);
 241             tz               = _mm_mul_ps(fscal,dz00);
 242
 243             /* Update vectorial force */
 244             fix0             = _mm_add_ps(fix0,tx);
 245             fiy0             = _mm_add_ps(fiy0,ty);
 246             fiz0             = _mm_add_ps(fiz0,tz);
 247
 248             fjptrA             = f+j_coord_offsetA;
 249             fjptrB             = f+j_coord_offsetB;
 250             fjptrC             = f+j_coord_offsetC;
 251             fjptrD             = f+j_coord_offsetD;
 252             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 253
 254             /* Inner loop uses 51 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             /* Get j neighbor index, and coordinate index */
 261             jnrlistA         = jjnr[jidx];
 262             jnrlistB         = jjnr[jidx+1];
 263             jnrlistC         = jjnr[jidx+2];
 264             jnrlistD         = jjnr[jidx+3];
 265             /* Sign of each element will be negative for non-real atoms.
 266              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 267              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 268              */
 269             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 270             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 271             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 272             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 273             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                               x+j_coord_offsetC,x+j_coord_offsetD,
 282                                               &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm_sub_ps(ix0,jx0);
 286             dy00             = _mm_sub_ps(iy0,jy0);
 287             dz00             = _mm_sub_ps(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 291
 292             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 293
 294             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 295
 296             /* Load parameters for j particles */
 297             vdwjidx0A        = 2*vdwtype[jnrA+0];
 298             vdwjidx0B        = 2*vdwtype[jnrB+0];
 299             vdwjidx0C        = 2*vdwtype[jnrC+0];
 300             vdwjidx0D        = 2*vdwtype[jnrD+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r00              = _mm_mul_ps(rsq00,rinv00);
 307             r00              = _mm_andnot_ps(dummy_mask,r00);
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 311                                          vdwparam+vdwioffset0+vdwjidx0B,
 312                                          vdwparam+vdwioffset0+vdwjidx0C,
 313                                          vdwparam+vdwioffset0+vdwjidx0D,
 314                                          &c6_00,&c12_00);
 315
 316             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 317                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 318                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 319                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 320
 321             /* Analytical LJ-PME */
 322             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 323             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 324             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 325             exponent         = gmx_simd_exp_r(ewcljrsq);
 326             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 327             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 328             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 329             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 330             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 331             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 332             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 333             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);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 337             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 338
 339             fscal            = fvdw;
 340
 341             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_ps(fscal,dx00);
 345             ty               = _mm_mul_ps(fscal,dy00);
 346             tz               = _mm_mul_ps(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm_add_ps(fix0,tx);
 350             fiy0             = _mm_add_ps(fiy0,ty);
 351             fiz0             = _mm_add_ps(fiz0,tz);
 352
 353             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 354             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 355             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 356             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 357             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 358
 359             /* Inner loop uses 52 flops */
 360         }
 361
 362         /* End of innermost loop */
 363
 364         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 365                                               f+i_coord_offset,fshift+i_shift_offset);
 366
 367         ggid                        = gid[iidx];
 368         /* Update potential energies */
 369         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 7 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
 386  * Electrostatics interaction: None
 387  * VdW interaction:            LJEwald
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB,jnrC,jnrD;
 409     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 410     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 412     real             rcutoff_scalar;
 413     real             *shiftvec,*fshift,*x,*f;
 414     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 415     real             scratch[4*DIM];
 416     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 417     int              vdwioffset0;
 418     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 419     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 420     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 421     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 422     int              nvdwtype;
 423     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 424     int              *vdwtype;
 425     real             *vdwparam;
 426     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 427     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 428     __m128           c6grid_00;
 429     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 430     real             *vdwgridparam;
 431     __m128           one_half  = _mm_set1_ps(0.5);
 432     __m128           minus_one = _mm_set1_ps(-1.0);
 433     __m128           dummy_mask,cutoff_mask;
 434     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 435     __m128           one     = _mm_set1_ps(1.0);
 436     __m128           two     = _mm_set1_ps(2.0);
 437     x                = xx[0];
 438     f                = ff[0];
 439
 440     nri              = nlist->nri;
 441     iinr             = nlist->iinr;
 442     jindex           = nlist->jindex;
 443     jjnr             = nlist->jjnr;
 444     shiftidx         = nlist->shift;
 445     gid              = nlist->gid;
 446     shiftvec         = fr->shift_vec[0];
 447     fshift           = fr->fshift[0];
 448     nvdwtype         = fr->ntype;
 449     vdwparam         = fr->nbfp;
 450     vdwtype          = mdatoms->typeA;
 451     vdwgridparam     = fr->ljpme_c6grid;
 452     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 453     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 454     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 455
 456     /* Avoid stupid compiler warnings */
 457     jnrA = jnrB = jnrC = jnrD = 0;
 458     j_coord_offsetA = 0;
 459     j_coord_offsetB = 0;
 460     j_coord_offsetC = 0;
 461     j_coord_offsetD = 0;
 462
 463     outeriter        = 0;
 464     inneriter        = 0;
 465
 466     for(iidx=0;iidx<4*DIM;iidx++)
 467     {
 468         scratch[iidx] = 0.0;
 469     }
 470
 471     /* Start outer loop over neighborlists */
 472     for(iidx=0; iidx<nri; iidx++)
 473     {
 474         /* Load shift vector for this list */
 475         i_shift_offset   = DIM*shiftidx[iidx];
 476
 477         /* Load limits for loop over neighbors */
 478         j_index_start    = jindex[iidx];
 479         j_index_end      = jindex[iidx+1];
 480
 481         /* Get outer coordinate index */
 482         inr              = iinr[iidx];
 483         i_coord_offset   = DIM*inr;
 484
 485         /* Load i particle coords and add shift vector */
 486         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 487
 488         fix0             = _mm_setzero_ps();
 489         fiy0             = _mm_setzero_ps();
 490         fiz0             = _mm_setzero_ps();
 491
 492         /* Load parameters for i particles */
 493         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 494
 495         /* Start inner kernel loop */
 496         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrA             = jjnr[jidx];
 501             jnrB             = jjnr[jidx+1];
 502             jnrC             = jjnr[jidx+2];
 503             jnrD             = jjnr[jidx+3];
 504             j_coord_offsetA  = DIM*jnrA;
 505             j_coord_offsetB  = DIM*jnrB;
 506             j_coord_offsetC  = DIM*jnrC;
 507             j_coord_offsetD  = DIM*jnrD;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               x+j_coord_offsetC,x+j_coord_offsetD,
 512                                               &jx0,&jy0,&jz0);
 513
 514             /* Calculate displacement vector */
 515             dx00             = _mm_sub_ps(ix0,jx0);
 516             dy00             = _mm_sub_ps(iy0,jy0);
 517             dz00             = _mm_sub_ps(iz0,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 521
 522             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 523
 524             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 525
 526             /* Load parameters for j particles */
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529             vdwjidx0C        = 2*vdwtype[jnrC+0];
 530             vdwjidx0D        = 2*vdwtype[jnrD+0];
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r00              = _mm_mul_ps(rsq00,rinv00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,
 541                                          vdwparam+vdwioffset0+vdwjidx0C,
 542                                          vdwparam+vdwioffset0+vdwjidx0D,
 543                                          &c6_00,&c12_00);
 544
 545             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 547                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 548                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 549
 550             /* Analytical LJ-PME */
 551             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 552             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 553             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 554             exponent         = gmx_simd_exp_r(ewcljrsq);
 555             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 556             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 557             /* f6A = 6 * C6grid * (1 - poly) */
 558             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 559             /* f6B = C6grid * exponent * beta^6 */
 560             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 561             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 562             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);
 563
 564             fscal            = fvdw;
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             fjptrA             = f+j_coord_offsetA;
 577             fjptrB             = f+j_coord_offsetB;
 578             fjptrC             = f+j_coord_offsetC;
 579             fjptrD             = f+j_coord_offsetD;
 580             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 581
 582             /* Inner loop uses 46 flops */
 583         }
 584
 585         if(jidx<j_index_end)
 586         {
 587
 588             /* Get j neighbor index, and coordinate index */
 589             jnrlistA         = jjnr[jidx];
 590             jnrlistB         = jjnr[jidx+1];
 591             jnrlistC         = jjnr[jidx+2];
 592             jnrlistD         = jjnr[jidx+3];
 593             /* Sign of each element will be negative for non-real atoms.
 594              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 595              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 596              */
 597             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 598             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 599             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 600             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 601             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 602             j_coord_offsetA  = DIM*jnrA;
 603             j_coord_offsetB  = DIM*jnrB;
 604             j_coord_offsetC  = DIM*jnrC;
 605             j_coord_offsetD  = DIM*jnrD;
 606
 607             /* load j atom coordinates */
 608             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 609                                               x+j_coord_offsetC,x+j_coord_offsetD,
 610                                               &jx0,&jy0,&jz0);
 611
 612             /* Calculate displacement vector */
 613             dx00             = _mm_sub_ps(ix0,jx0);
 614             dy00             = _mm_sub_ps(iy0,jy0);
 615             dz00             = _mm_sub_ps(iz0,jz0);
 616
 617             /* Calculate squared distance and things based on it */
 618             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 619
 620             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 621
 622             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 623
 624             /* Load parameters for j particles */
 625             vdwjidx0A        = 2*vdwtype[jnrA+0];
 626             vdwjidx0B        = 2*vdwtype[jnrB+0];
 627             vdwjidx0C        = 2*vdwtype[jnrC+0];
 628             vdwjidx0D        = 2*vdwtype[jnrD+0];
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r00              = _mm_mul_ps(rsq00,rinv00);
 635             r00              = _mm_andnot_ps(dummy_mask,r00);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 639                                          vdwparam+vdwioffset0+vdwjidx0B,
 640                                          vdwparam+vdwioffset0+vdwjidx0C,
 641                                          vdwparam+vdwioffset0+vdwjidx0D,
 642                                          &c6_00,&c12_00);
 643
 644             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 645                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 646                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 647                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 648
 649             /* Analytical LJ-PME */
 650             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 651             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 652             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 653             exponent         = gmx_simd_exp_r(ewcljrsq);
 654             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 655             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 656             /* f6A = 6 * C6grid * (1 - poly) */
 657             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 658             /* f6B = C6grid * exponent * beta^6 */
 659             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 660             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 661             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);
 662
 663             fscal            = fvdw;
 664
 665             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm_mul_ps(fscal,dx00);
 669             ty               = _mm_mul_ps(fscal,dy00);
 670             tz               = _mm_mul_ps(fscal,dz00);
 671
 672             /* Update vectorial force */
 673             fix0             = _mm_add_ps(fix0,tx);
 674             fiy0             = _mm_add_ps(fiy0,ty);
 675             fiz0             = _mm_add_ps(fiz0,tz);
 676
 677             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 678             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 679             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 680             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 681             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 682
 683             /* Inner loop uses 47 flops */
 684         }
 685
 686         /* End of innermost loop */
 687
 688         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 689                                               f+i_coord_offset,fshift+i_shift_offset);
 690
 691         /* Increment number of inner iterations */
 692         inneriter                  += j_index_end - j_index_start;
 693
 694         /* Outer loop uses 6 flops */
 695     }
 696
 697     /* Increment number of outer iterations */
 698     outeriter        += nri;
 699
 700     /* Update outer/inner flops */
 701
 702     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 703 }