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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           c6grid_00;
  97     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  98     real             *vdwgridparam;
  99     __m128           one_half  = _mm_set1_ps(0.5);
 100     __m128           minus_one = _mm_set1_ps(-1.0);
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119     vdwgridparam     = fr->ljpme_c6grid;
 120     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 121     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 122     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159
 160         /* Load parameters for i particles */
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         vvdwsum          = _mm_setzero_ps();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               x+j_coord_offsetC,x+j_coord_offsetD,
 183                                               &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm_sub_ps(ix0,jx0);
 187             dy00             = _mm_sub_ps(iy0,jy0);
 188             dz00             = _mm_sub_ps(iz0,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 192
 193             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 194
 195             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 196
 197             /* Load parameters for j particles */
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200             vdwjidx0C        = 2*vdwtype[jnrC+0];
 201             vdwjidx0D        = 2*vdwtype[jnrD+0];
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             r00              = _mm_mul_ps(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 211                                          vdwparam+vdwioffset0+vdwjidx0B,
 212                                          vdwparam+vdwioffset0+vdwjidx0C,
 213                                          vdwparam+vdwioffset0+vdwjidx0D,
 214                                          &c6_00,&c12_00);
 215
 216             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 217                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 218                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 219                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 220
 221             /* Analytical LJ-PME */
 222             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 223             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 224             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 225             exponent         = gmx_simd_exp_r(ewcljrsq);
 226             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 227             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 228             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 229             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 230             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 231             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 232             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 233             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);
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 237
 238             fscal            = fvdw;
 239
 240             /* Calculate temporary vectorial force */
 241             tx               = _mm_mul_ps(fscal,dx00);
 242             ty               = _mm_mul_ps(fscal,dy00);
 243             tz               = _mm_mul_ps(fscal,dz00);
 244
 245             /* Update vectorial force */
 246             fix0             = _mm_add_ps(fix0,tx);
 247             fiy0             = _mm_add_ps(fiy0,ty);
 248             fiz0             = _mm_add_ps(fiz0,tz);
 249
 250             fjptrA             = f+j_coord_offsetA;
 251             fjptrB             = f+j_coord_offsetB;
 252             fjptrC             = f+j_coord_offsetC;
 253             fjptrD             = f+j_coord_offsetD;
 254             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 255
 256             /* Inner loop uses 51 flops */
 257         }
 258
 259         if(jidx<j_index_end)
 260         {
 261
 262             /* Get j neighbor index, and coordinate index */
 263             jnrlistA         = jjnr[jidx];
 264             jnrlistB         = jjnr[jidx+1];
 265             jnrlistC         = jjnr[jidx+2];
 266             jnrlistD         = jjnr[jidx+3];
 267             /* Sign of each element will be negative for non-real atoms.
 268              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 269              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 270              */
 271             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 272             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 273             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 274             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 275             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 276             j_coord_offsetA  = DIM*jnrA;
 277             j_coord_offsetB  = DIM*jnrB;
 278             j_coord_offsetC  = DIM*jnrC;
 279             j_coord_offsetD  = DIM*jnrD;
 280
 281             /* load j atom coordinates */
 282             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 283                                               x+j_coord_offsetC,x+j_coord_offsetD,
 284                                               &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _mm_sub_ps(ix0,jx0);
 288             dy00             = _mm_sub_ps(iy0,jy0);
 289             dz00             = _mm_sub_ps(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 295
 296             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             vdwjidx0A        = 2*vdwtype[jnrA+0];
 300             vdwjidx0B        = 2*vdwtype[jnrB+0];
 301             vdwjidx0C        = 2*vdwtype[jnrC+0];
 302             vdwjidx0D        = 2*vdwtype[jnrD+0];
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r00              = _mm_mul_ps(rsq00,rinv00);
 309             r00              = _mm_andnot_ps(dummy_mask,r00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 313                                          vdwparam+vdwioffset0+vdwjidx0B,
 314                                          vdwparam+vdwioffset0+vdwjidx0C,
 315                                          vdwparam+vdwioffset0+vdwjidx0D,
 316                                          &c6_00,&c12_00);
 317
 318             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 319                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 320                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 321                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 322
 323             /* Analytical LJ-PME */
 324             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 326             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 327             exponent         = gmx_simd_exp_r(ewcljrsq);
 328             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 329             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 330             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 331             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 332             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 333             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 334             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 335             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);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 339             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = fvdw;
 342
 343             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_ps(fscal,dx00);
 347             ty               = _mm_mul_ps(fscal,dy00);
 348             tz               = _mm_mul_ps(fscal,dz00);
 349
 350             /* Update vectorial force */
 351             fix0             = _mm_add_ps(fix0,tx);
 352             fiy0             = _mm_add_ps(fiy0,ty);
 353             fiz0             = _mm_add_ps(fiz0,tz);
 354
 355             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 356             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 357             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 358             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 359             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 360
 361             /* Inner loop uses 52 flops */
 362         }
 363
 364         /* End of innermost loop */
 365
 366         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 367                                               f+i_coord_offset,fshift+i_shift_offset);
 368
 369         ggid                        = gid[iidx];
 370         /* Update potential energies */
 371         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 372
 373         /* Increment number of inner iterations */
 374         inneriter                  += j_index_end - j_index_start;
 375
 376         /* Outer loop uses 7 flops */
 377     }
 378
 379     /* Increment number of outer iterations */
 380     outeriter        += nri;
 381
 382     /* Update outer/inner flops */
 383
 384     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
 385 }
 386 /*
 387  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
 388  * Electrostatics interaction: None
 389  * VdW interaction:            LJEwald
 390  * Geometry:                   Particle-Particle
 391  * Calculate force/pot:        Force
 392  */
 393 void
 394 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
 395                     (t_nblist                    * gmx_restrict       nlist,
 396                      rvec                        * gmx_restrict          xx,
 397                      rvec                        * gmx_restrict          ff,
 398                      t_forcerec                  * gmx_restrict          fr,
 399                      t_mdatoms                   * gmx_restrict     mdatoms,
 400                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 401                      t_nrnb                      * gmx_restrict        nrnb)
 402 {
 403     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 404      * just 0 for non-waters.
 405      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 406      * jnr indices corresponding to data put in the four positions in the SIMD register.
 407      */
 408     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 409     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 410     int              jnrA,jnrB,jnrC,jnrD;
 411     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 412     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 413     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 414     real             rcutoff_scalar;
 415     real             *shiftvec,*fshift,*x,*f;
 416     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 417     real             scratch[4*DIM];
 418     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 422     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     int              nvdwtype;
 425     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 426     int              *vdwtype;
 427     real             *vdwparam;
 428     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 429     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 430     __m128           c6grid_00;
 431     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 432     real             *vdwgridparam;
 433     __m128           one_half  = _mm_set1_ps(0.5);
 434     __m128           minus_one = _mm_set1_ps(-1.0);
 435     __m128           dummy_mask,cutoff_mask;
 436     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 437     __m128           one     = _mm_set1_ps(1.0);
 438     __m128           two     = _mm_set1_ps(2.0);
 439     x                = xx[0];
 440     f                = ff[0];
 441
 442     nri              = nlist->nri;
 443     iinr             = nlist->iinr;
 444     jindex           = nlist->jindex;
 445     jjnr             = nlist->jjnr;
 446     shiftidx         = nlist->shift;
 447     gid              = nlist->gid;
 448     shiftvec         = fr->shift_vec[0];
 449     fshift           = fr->fshift[0];
 450     nvdwtype         = fr->ntype;
 451     vdwparam         = fr->nbfp;
 452     vdwtype          = mdatoms->typeA;
 453     vdwgridparam     = fr->ljpme_c6grid;
 454     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 455     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 456     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 457
 458     /* Avoid stupid compiler warnings */
 459     jnrA = jnrB = jnrC = jnrD = 0;
 460     j_coord_offsetA = 0;
 461     j_coord_offsetB = 0;
 462     j_coord_offsetC = 0;
 463     j_coord_offsetD = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     for(iidx=0;iidx<4*DIM;iidx++)
 469     {
 470         scratch[iidx] = 0.0;
 471     }
 472
 473     /* Start outer loop over neighborlists */
 474     for(iidx=0; iidx<nri; iidx++)
 475     {
 476         /* Load shift vector for this list */
 477         i_shift_offset   = DIM*shiftidx[iidx];
 478
 479         /* Load limits for loop over neighbors */
 480         j_index_start    = jindex[iidx];
 481         j_index_end      = jindex[iidx+1];
 482
 483         /* Get outer coordinate index */
 484         inr              = iinr[iidx];
 485         i_coord_offset   = DIM*inr;
 486
 487         /* Load i particle coords and add shift vector */
 488         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 489
 490         fix0             = _mm_setzero_ps();
 491         fiy0             = _mm_setzero_ps();
 492         fiz0             = _mm_setzero_ps();
 493
 494         /* Load parameters for i particles */
 495         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 496
 497         /* Start inner kernel loop */
 498         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrA             = jjnr[jidx];
 503             jnrB             = jjnr[jidx+1];
 504             jnrC             = jjnr[jidx+2];
 505             jnrD             = jjnr[jidx+3];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508             j_coord_offsetC  = DIM*jnrC;
 509             j_coord_offsetD  = DIM*jnrD;
 510
 511             /* load j atom coordinates */
 512             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               x+j_coord_offsetC,x+j_coord_offsetD,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _mm_sub_ps(ix0,jx0);
 518             dy00             = _mm_sub_ps(iy0,jy0);
 519             dz00             = _mm_sub_ps(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 523
 524             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 525
 526             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 527
 528             /* Load parameters for j particles */
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531             vdwjidx0C        = 2*vdwtype[jnrC+0];
 532             vdwjidx0D        = 2*vdwtype[jnrD+0];
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r00              = _mm_mul_ps(rsq00,rinv00);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 542                                          vdwparam+vdwioffset0+vdwjidx0B,
 543                                          vdwparam+vdwioffset0+vdwjidx0C,
 544                                          vdwparam+vdwioffset0+vdwjidx0D,
 545                                          &c6_00,&c12_00);
 546
 547             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 548                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 549                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 550                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 551
 552             /* Analytical LJ-PME */
 553             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 554             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 555             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 556             exponent         = gmx_simd_exp_r(ewcljrsq);
 557             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 558             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 559             /* f6A = 6 * C6grid * (1 - poly) */
 560             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 561             /* f6B = C6grid * exponent * beta^6 */
 562             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 563             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 564             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);
 565
 566             fscal            = fvdw;
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm_mul_ps(fscal,dx00);
 570             ty               = _mm_mul_ps(fscal,dy00);
 571             tz               = _mm_mul_ps(fscal,dz00);
 572
 573             /* Update vectorial force */
 574             fix0             = _mm_add_ps(fix0,tx);
 575             fiy0             = _mm_add_ps(fiy0,ty);
 576             fiz0             = _mm_add_ps(fiz0,tz);
 577
 578             fjptrA             = f+j_coord_offsetA;
 579             fjptrB             = f+j_coord_offsetB;
 580             fjptrC             = f+j_coord_offsetC;
 581             fjptrD             = f+j_coord_offsetD;
 582             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 583
 584             /* Inner loop uses 46 flops */
 585         }
 586
 587         if(jidx<j_index_end)
 588         {
 589
 590             /* Get j neighbor index, and coordinate index */
 591             jnrlistA         = jjnr[jidx];
 592             jnrlistB         = jjnr[jidx+1];
 593             jnrlistC         = jjnr[jidx+2];
 594             jnrlistD         = jjnr[jidx+3];
 595             /* Sign of each element will be negative for non-real atoms.
 596              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 597              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 598              */
 599             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 600             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 601             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 602             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 603             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 604             j_coord_offsetA  = DIM*jnrA;
 605             j_coord_offsetB  = DIM*jnrB;
 606             j_coord_offsetC  = DIM*jnrC;
 607             j_coord_offsetD  = DIM*jnrD;
 608
 609             /* load j atom coordinates */
 610             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 611                                               x+j_coord_offsetC,x+j_coord_offsetD,
 612                                               &jx0,&jy0,&jz0);
 613
 614             /* Calculate displacement vector */
 615             dx00             = _mm_sub_ps(ix0,jx0);
 616             dy00             = _mm_sub_ps(iy0,jy0);
 617             dz00             = _mm_sub_ps(iz0,jz0);
 618
 619             /* Calculate squared distance and things based on it */
 620             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 621
 622             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 623
 624             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 625
 626             /* Load parameters for j particles */
 627             vdwjidx0A        = 2*vdwtype[jnrA+0];
 628             vdwjidx0B        = 2*vdwtype[jnrB+0];
 629             vdwjidx0C        = 2*vdwtype[jnrC+0];
 630             vdwjidx0D        = 2*vdwtype[jnrD+0];
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             r00              = _mm_mul_ps(rsq00,rinv00);
 637             r00              = _mm_andnot_ps(dummy_mask,r00);
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 641                                          vdwparam+vdwioffset0+vdwjidx0B,
 642                                          vdwparam+vdwioffset0+vdwjidx0C,
 643                                          vdwparam+vdwioffset0+vdwjidx0D,
 644                                          &c6_00,&c12_00);
 645
 646             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 647                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 648                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 649                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 650
 651             /* Analytical LJ-PME */
 652             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 653             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 654             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 655             exponent         = gmx_simd_exp_r(ewcljrsq);
 656             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 657             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 658             /* f6A = 6 * C6grid * (1 - poly) */
 659             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 660             /* f6B = C6grid * exponent * beta^6 */
 661             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 662             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 663             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);
 664
 665             fscal            = fvdw;
 666
 667             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 668
 669             /* Calculate temporary vectorial force */
 670             tx               = _mm_mul_ps(fscal,dx00);
 671             ty               = _mm_mul_ps(fscal,dy00);
 672             tz               = _mm_mul_ps(fscal,dz00);
 673
 674             /* Update vectorial force */
 675             fix0             = _mm_add_ps(fix0,tx);
 676             fiy0             = _mm_add_ps(fiy0,ty);
 677             fiz0             = _mm_add_ps(fiz0,tz);
 678
 679             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 680             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 681             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 682             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 683             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 684
 685             /* Inner loop uses 47 flops */
 686         }
 687
 688         /* End of innermost loop */
 689
 690         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 691                                               f+i_coord_offset,fshift+i_shift_offset);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 6 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 705 }