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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  92     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  93     __m128           c6grid_00;
  94     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     real             *vdwgridparam;
  96     __m128           one_half = _mm_set1_ps(0.5);
  97     __m128           minus_one = _mm_set1_ps(-1.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116     vdwgridparam     = fr->ljpme_c6grid;
 117     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 118     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 119     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_ps();
 154         fiy0             = _mm_setzero_ps();
 155         fiz0             = _mm_setzero_ps();
 156
 157         /* Load parameters for i particles */
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         vvdwsum          = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 189
 190             rinv00           = sse2_invsqrt_f(rsq00);
 191
 192             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197             vdwjidx0C        = 2*vdwtype[jnrC+0];
 198             vdwjidx0D        = 2*vdwtype[jnrD+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             r00              = _mm_mul_ps(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,
 209                                          vdwparam+vdwioffset0+vdwjidx0C,
 210                                          vdwparam+vdwioffset0+vdwjidx0D,
 211                                          &c6_00,&c12_00);
 212             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 213                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 214                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 215                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 216
 217             /* Analytical LJ-PME */
 218             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 219             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 220             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 221             exponent         = sse2_exp_f(ewcljrsq);
 222             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 223             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 224             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 225             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 226             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 227             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 228             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 229             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);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 233
 234             fscal            = fvdw;
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_ps(fscal,dx00);
 238             ty               = _mm_mul_ps(fscal,dy00);
 239             tz               = _mm_mul_ps(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_ps(fix0,tx);
 243             fiy0             = _mm_add_ps(fiy0,ty);
 244             fiz0             = _mm_add_ps(fiz0,tz);
 245
 246             fjptrA             = f+j_coord_offsetA;
 247             fjptrB             = f+j_coord_offsetB;
 248             fjptrC             = f+j_coord_offsetC;
 249             fjptrD             = f+j_coord_offsetD;
 250             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 251
 252             /* Inner loop uses 51 flops */
 253         }
 254
 255         if(jidx<j_index_end)
 256         {
 257
 258             /* Get j neighbor index, and coordinate index */
 259             jnrlistA         = jjnr[jidx];
 260             jnrlistB         = jjnr[jidx+1];
 261             jnrlistC         = jjnr[jidx+2];
 262             jnrlistD         = jjnr[jidx+3];
 263             /* Sign of each element will be negative for non-real atoms.
 264              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 265              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 266              */
 267             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276
 277             /* load j atom coordinates */
 278             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 279                                               x+j_coord_offsetC,x+j_coord_offsetD,
 280                                               &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm_sub_ps(ix0,jx0);
 284             dy00             = _mm_sub_ps(iy0,jy0);
 285             dz00             = _mm_sub_ps(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 289
 290             rinv00           = sse2_invsqrt_f(rsq00);
 291
 292             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 293
 294             /* Load parameters for j particles */
 295             vdwjidx0A        = 2*vdwtype[jnrA+0];
 296             vdwjidx0B        = 2*vdwtype[jnrB+0];
 297             vdwjidx0C        = 2*vdwtype[jnrC+0];
 298             vdwjidx0D        = 2*vdwtype[jnrD+0];
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r00              = _mm_mul_ps(rsq00,rinv00);
 305             r00              = _mm_andnot_ps(dummy_mask,r00);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 309                                          vdwparam+vdwioffset0+vdwjidx0B,
 310                                          vdwparam+vdwioffset0+vdwjidx0C,
 311                                          vdwparam+vdwioffset0+vdwjidx0D,
 312                                          &c6_00,&c12_00);
 313             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 314                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 315                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 316                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 317
 318             /* Analytical LJ-PME */
 319             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 320             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 321             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 322             exponent         = sse2_exp_f(ewcljrsq);
 323             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 324             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 325             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 326             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 327             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 328             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 329             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 330             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);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 334             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 335
 336             fscal            = fvdw;
 337
 338             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm_mul_ps(fscal,dx00);
 342             ty               = _mm_mul_ps(fscal,dy00);
 343             tz               = _mm_mul_ps(fscal,dz00);
 344
 345             /* Update vectorial force */
 346             fix0             = _mm_add_ps(fix0,tx);
 347             fiy0             = _mm_add_ps(fiy0,ty);
 348             fiz0             = _mm_add_ps(fiz0,tz);
 349
 350             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 351             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 352             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 353             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 354             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 355
 356             /* Inner loop uses 52 flops */
 357         }
 358
 359         /* End of innermost loop */
 360
 361         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 362                                               f+i_coord_offset,fshift+i_shift_offset);
 363
 364         ggid                        = gid[iidx];
 365         /* Update potential energies */
 366         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 367
 368         /* Increment number of inner iterations */
 369         inneriter                  += j_index_end - j_index_start;
 370
 371         /* Outer loop uses 7 flops */
 372     }
 373
 374     /* Increment number of outer iterations */
 375     outeriter        += nri;
 376
 377     /* Update outer/inner flops */
 378
 379     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
 380 }
 381 /*
 382  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_single
 383  * Electrostatics interaction: None
 384  * VdW interaction:            LJEwald
 385  * Geometry:                   Particle-Particle
 386  * Calculate force/pot:        Force
 387  */
 388 void
 389 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_single
 390                     (t_nblist                    * gmx_restrict       nlist,
 391                      rvec                        * gmx_restrict          xx,
 392                      rvec                        * gmx_restrict          ff,
 393                      struct t_forcerec           * gmx_restrict          fr,
 394                      t_mdatoms                   * gmx_restrict     mdatoms,
 395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 396                      t_nrnb                      * gmx_restrict        nrnb)
 397 {
 398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 399      * just 0 for non-waters.
 400      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 401      * jnr indices corresponding to data put in the four positions in the SIMD register.
 402      */
 403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 405     int              jnrA,jnrB,jnrC,jnrD;
 406     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 407     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 409     real             rcutoff_scalar;
 410     real             *shiftvec,*fshift,*x,*f;
 411     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 412     real             scratch[4*DIM];
 413     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     int              vdwioffset0;
 415     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 417     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 419     int              nvdwtype;
 420     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 421     int              *vdwtype;
 422     real             *vdwparam;
 423     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 424     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 425     __m128           c6grid_00;
 426     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 427     real             *vdwgridparam;
 428     __m128           one_half = _mm_set1_ps(0.5);
 429     __m128           minus_one = _mm_set1_ps(-1.0);
 430     __m128           dummy_mask,cutoff_mask;
 431     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 432     __m128           one     = _mm_set1_ps(1.0);
 433     __m128           two     = _mm_set1_ps(2.0);
 434     x                = xx[0];
 435     f                = ff[0];
 436
 437     nri              = nlist->nri;
 438     iinr             = nlist->iinr;
 439     jindex           = nlist->jindex;
 440     jjnr             = nlist->jjnr;
 441     shiftidx         = nlist->shift;
 442     gid              = nlist->gid;
 443     shiftvec         = fr->shift_vec[0];
 444     fshift           = fr->fshift[0];
 445     nvdwtype         = fr->ntype;
 446     vdwparam         = fr->nbfp;
 447     vdwtype          = mdatoms->typeA;
 448     vdwgridparam     = fr->ljpme_c6grid;
 449     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 450     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 451     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 452
 453     /* Avoid stupid compiler warnings */
 454     jnrA = jnrB = jnrC = jnrD = 0;
 455     j_coord_offsetA = 0;
 456     j_coord_offsetB = 0;
 457     j_coord_offsetC = 0;
 458     j_coord_offsetD = 0;
 459
 460     outeriter        = 0;
 461     inneriter        = 0;
 462
 463     for(iidx=0;iidx<4*DIM;iidx++)
 464     {
 465         scratch[iidx] = 0.0;
 466     }
 467
 468     /* Start outer loop over neighborlists */
 469     for(iidx=0; iidx<nri; iidx++)
 470     {
 471         /* Load shift vector for this list */
 472         i_shift_offset   = DIM*shiftidx[iidx];
 473
 474         /* Load limits for loop over neighbors */
 475         j_index_start    = jindex[iidx];
 476         j_index_end      = jindex[iidx+1];
 477
 478         /* Get outer coordinate index */
 479         inr              = iinr[iidx];
 480         i_coord_offset   = DIM*inr;
 481
 482         /* Load i particle coords and add shift vector */
 483         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 484
 485         fix0             = _mm_setzero_ps();
 486         fiy0             = _mm_setzero_ps();
 487         fiz0             = _mm_setzero_ps();
 488
 489         /* Load parameters for i particles */
 490         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 491
 492         /* Start inner kernel loop */
 493         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 494         {
 495
 496             /* Get j neighbor index, and coordinate index */
 497             jnrA             = jjnr[jidx];
 498             jnrB             = jjnr[jidx+1];
 499             jnrC             = jjnr[jidx+2];
 500             jnrD             = jjnr[jidx+3];
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505
 506             /* load j atom coordinates */
 507             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 508                                               x+j_coord_offsetC,x+j_coord_offsetD,
 509                                               &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm_sub_ps(ix0,jx0);
 513             dy00             = _mm_sub_ps(iy0,jy0);
 514             dz00             = _mm_sub_ps(iz0,jz0);
 515
 516             /* Calculate squared distance and things based on it */
 517             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 518
 519             rinv00           = sse2_invsqrt_f(rsq00);
 520
 521             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 522
 523             /* Load parameters for j particles */
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526             vdwjidx0C        = 2*vdwtype[jnrC+0];
 527             vdwjidx0D        = 2*vdwtype[jnrD+0];
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r00              = _mm_mul_ps(rsq00,rinv00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,
 538                                          vdwparam+vdwioffset0+vdwjidx0C,
 539                                          vdwparam+vdwioffset0+vdwjidx0D,
 540                                          &c6_00,&c12_00);
 541             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 542                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 545
 546             /* Analytical LJ-PME */
 547             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 548             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 549             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 550             exponent         = sse2_exp_f(ewcljrsq);
 551             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 552             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 553             /* f6A = 6 * C6grid * (1 - poly) */
 554             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 555             /* f6B = C6grid * exponent * beta^6 */
 556             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 557             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 558             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);
 559
 560             fscal            = fvdw;
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_ps(fscal,dx00);
 564             ty               = _mm_mul_ps(fscal,dy00);
 565             tz               = _mm_mul_ps(fscal,dz00);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm_add_ps(fix0,tx);
 569             fiy0             = _mm_add_ps(fiy0,ty);
 570             fiz0             = _mm_add_ps(fiz0,tz);
 571
 572             fjptrA             = f+j_coord_offsetA;
 573             fjptrB             = f+j_coord_offsetB;
 574             fjptrC             = f+j_coord_offsetC;
 575             fjptrD             = f+j_coord_offsetD;
 576             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 577
 578             /* Inner loop uses 46 flops */
 579         }
 580
 581         if(jidx<j_index_end)
 582         {
 583
 584             /* Get j neighbor index, and coordinate index */
 585             jnrlistA         = jjnr[jidx];
 586             jnrlistB         = jjnr[jidx+1];
 587             jnrlistC         = jjnr[jidx+2];
 588             jnrlistD         = jjnr[jidx+3];
 589             /* Sign of each element will be negative for non-real atoms.
 590              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 591              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 592              */
 593             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 594             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 595             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 596             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 597             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 598             j_coord_offsetA  = DIM*jnrA;
 599             j_coord_offsetB  = DIM*jnrB;
 600             j_coord_offsetC  = DIM*jnrC;
 601             j_coord_offsetD  = DIM*jnrD;
 602
 603             /* load j atom coordinates */
 604             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 605                                               x+j_coord_offsetC,x+j_coord_offsetD,
 606                                               &jx0,&jy0,&jz0);
 607
 608             /* Calculate displacement vector */
 609             dx00             = _mm_sub_ps(ix0,jx0);
 610             dy00             = _mm_sub_ps(iy0,jy0);
 611             dz00             = _mm_sub_ps(iz0,jz0);
 612
 613             /* Calculate squared distance and things based on it */
 614             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 615
 616             rinv00           = sse2_invsqrt_f(rsq00);
 617
 618             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 619
 620             /* Load parameters for j particles */
 621             vdwjidx0A        = 2*vdwtype[jnrA+0];
 622             vdwjidx0B        = 2*vdwtype[jnrB+0];
 623             vdwjidx0C        = 2*vdwtype[jnrC+0];
 624             vdwjidx0D        = 2*vdwtype[jnrD+0];
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r00              = _mm_mul_ps(rsq00,rinv00);
 631             r00              = _mm_andnot_ps(dummy_mask,r00);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 635                                          vdwparam+vdwioffset0+vdwjidx0B,
 636                                          vdwparam+vdwioffset0+vdwjidx0C,
 637                                          vdwparam+vdwioffset0+vdwjidx0D,
 638                                          &c6_00,&c12_00);
 639             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 640                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 641                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 642                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 643
 644             /* Analytical LJ-PME */
 645             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 646             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 647             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 648             exponent         = sse2_exp_f(ewcljrsq);
 649             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 650             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 651             /* f6A = 6 * C6grid * (1 - poly) */
 652             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 653             /* f6B = C6grid * exponent * beta^6 */
 654             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 655             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 656             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);
 657
 658             fscal            = fvdw;
 659
 660             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_ps(fscal,dx00);
 664             ty               = _mm_mul_ps(fscal,dy00);
 665             tz               = _mm_mul_ps(fscal,dz00);
 666
 667             /* Update vectorial force */
 668             fix0             = _mm_add_ps(fix0,tx);
 669             fiy0             = _mm_add_ps(fiy0,ty);
 670             fiz0             = _mm_add_ps(fiz0,tz);
 671
 672             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 673             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 674             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 675             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 676             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 677
 678             /* Inner loop uses 47 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         /* Increment number of inner iterations */
 687         inneriter                  += j_index_end - j_index_start;
 688
 689         /* Outer loop uses 6 flops */
 690     }
 691
 692     /* Increment number of outer iterations */
 693     outeriter        += nri;
 694
 695     /* Update outer/inner flops */
 696
 697     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 698 }