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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     int              nvdwtype;
  89     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  93     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  94     __m128           c6grid_00;
  95     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  96     real             *vdwgridparam;
  97     __m128           one_half = _mm_set1_ps(0.5);
  98     __m128           minus_one = _mm_set1_ps(-1.0);
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117     vdwgridparam     = fr->ljpme_c6grid;
 118     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 119     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 120     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 121
 122     rcutoff_scalar   = fr->rvdw;
 123     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 127     rvdw             = _mm_set1_ps(fr->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 167
 168         /* Reset potential sums */
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_ps(ix0,jx0);
 192             dy00             = _mm_sub_ps(iy0,jy0);
 193             dz00             = _mm_sub_ps(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 199
 200             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             vdwjidx0A        = 2*vdwtype[jnrA+0];
 204             vdwjidx0B        = 2*vdwtype[jnrB+0];
 205             vdwjidx0C        = 2*vdwtype[jnrC+0];
 206             vdwjidx0D        = 2*vdwtype[jnrD+0];
 207
 208             /**************************
 209              * CALCULATE INTERACTIONS *
 210              **************************/
 211
 212             if (gmx_mm_any_lt(rsq00,rcutoff2))
 213             {
 214
 215             r00              = _mm_mul_ps(rsq00,rinv00);
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,
 220                                          vdwparam+vdwioffset0+vdwjidx0C,
 221                                          vdwparam+vdwioffset0+vdwjidx0D,
 222                                          &c6_00,&c12_00);
 223             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 224                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 225                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 226                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 227
 228             /* Analytical LJ-PME */
 229             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 230             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 231             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 232             exponent         = gmx_simd_exp_r(ewcljrsq);
 233             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 234             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 235             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 236             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 237             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 238             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 239                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 240             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 241             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);
 242
 243             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 248
 249             fscal            = fvdw;
 250
 251             fscal            = _mm_and_ps(fscal,cutoff_mask);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             fjptrA             = f+j_coord_offsetA;
 264             fjptrB             = f+j_coord_offsetB;
 265             fjptrC             = f+j_coord_offsetC;
 266             fjptrD             = f+j_coord_offsetD;
 267             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 268
 269             }
 270
 271             /* Inner loop uses 62 flops */
 272         }
 273
 274         if(jidx<j_index_end)
 275         {
 276
 277             /* Get j neighbor index, and coordinate index */
 278             jnrlistA         = jjnr[jidx];
 279             jnrlistB         = jjnr[jidx+1];
 280             jnrlistC         = jjnr[jidx+2];
 281             jnrlistD         = jjnr[jidx+3];
 282             /* Sign of each element will be negative for non-real atoms.
 283              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 284              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 285              */
 286             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 287             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 288             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 289             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 290             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 291             j_coord_offsetA  = DIM*jnrA;
 292             j_coord_offsetB  = DIM*jnrB;
 293             j_coord_offsetC  = DIM*jnrC;
 294             j_coord_offsetD  = DIM*jnrD;
 295
 296             /* load j atom coordinates */
 297             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 298                                               x+j_coord_offsetC,x+j_coord_offsetD,
 299                                               &jx0,&jy0,&jz0);
 300
 301             /* Calculate displacement vector */
 302             dx00             = _mm_sub_ps(ix0,jx0);
 303             dy00             = _mm_sub_ps(iy0,jy0);
 304             dz00             = _mm_sub_ps(iz0,jz0);
 305
 306             /* Calculate squared distance and things based on it */
 307             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 308
 309             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 310
 311             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 312
 313             /* Load parameters for j particles */
 314             vdwjidx0A        = 2*vdwtype[jnrA+0];
 315             vdwjidx0B        = 2*vdwtype[jnrB+0];
 316             vdwjidx0C        = 2*vdwtype[jnrC+0];
 317             vdwjidx0D        = 2*vdwtype[jnrD+0];
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq00,rcutoff2))
 324             {
 325
 326             r00              = _mm_mul_ps(rsq00,rinv00);
 327             r00              = _mm_andnot_ps(dummy_mask,r00);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 331                                          vdwparam+vdwioffset0+vdwjidx0B,
 332                                          vdwparam+vdwioffset0+vdwjidx0C,
 333                                          vdwparam+vdwioffset0+vdwjidx0D,
 334                                          &c6_00,&c12_00);
 335             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 336                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 337                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 338                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 339
 340             /* Analytical LJ-PME */
 341             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 342             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 343             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 344             exponent         = gmx_simd_exp_r(ewcljrsq);
 345             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 346             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 347             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 348             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 349             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 350             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 351                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 352             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 353             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);
 354
 355             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 359             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 360             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 361
 362             fscal            = fvdw;
 363
 364             fscal            = _mm_and_ps(fscal,cutoff_mask);
 365
 366             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm_mul_ps(fscal,dx00);
 370             ty               = _mm_mul_ps(fscal,dy00);
 371             tz               = _mm_mul_ps(fscal,dz00);
 372
 373             /* Update vectorial force */
 374             fix0             = _mm_add_ps(fix0,tx);
 375             fiy0             = _mm_add_ps(fiy0,ty);
 376             fiz0             = _mm_add_ps(fiz0,tz);
 377
 378             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 379             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 380             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 381             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 382             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 383
 384             }
 385
 386             /* Inner loop uses 63 flops */
 387         }
 388
 389         /* End of innermost loop */
 390
 391         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 392                                               f+i_coord_offset,fshift+i_shift_offset);
 393
 394         ggid                        = gid[iidx];
 395         /* Update potential energies */
 396         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 397
 398         /* Increment number of inner iterations */
 399         inneriter                  += j_index_end - j_index_start;
 400
 401         /* Outer loop uses 7 flops */
 402     }
 403
 404     /* Increment number of outer iterations */
 405     outeriter        += nri;
 406
 407     /* Update outer/inner flops */
 408
 409     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*63);
 410 }
 411 /*
 412  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_single
 413  * Electrostatics interaction: None
 414  * VdW interaction:            LJEwald
 415  * Geometry:                   Particle-Particle
 416  * Calculate force/pot:        Force
 417  */
 418 void
 419 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_single
 420                     (t_nblist                    * gmx_restrict       nlist,
 421                      rvec                        * gmx_restrict          xx,
 422                      rvec                        * gmx_restrict          ff,
 423                      t_forcerec                  * gmx_restrict          fr,
 424                      t_mdatoms                   * gmx_restrict     mdatoms,
 425                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 426                      t_nrnb                      * gmx_restrict        nrnb)
 427 {
 428     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 429      * just 0 for non-waters.
 430      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 431      * jnr indices corresponding to data put in the four positions in the SIMD register.
 432      */
 433     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 434     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 435     int              jnrA,jnrB,jnrC,jnrD;
 436     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 437     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 438     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 439     real             rcutoff_scalar;
 440     real             *shiftvec,*fshift,*x,*f;
 441     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 442     real             scratch[4*DIM];
 443     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 444     int              vdwioffset0;
 445     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 446     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 447     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 448     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 449     int              nvdwtype;
 450     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 451     int              *vdwtype;
 452     real             *vdwparam;
 453     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 454     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 455     __m128           c6grid_00;
 456     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 457     real             *vdwgridparam;
 458     __m128           one_half = _mm_set1_ps(0.5);
 459     __m128           minus_one = _mm_set1_ps(-1.0);
 460     __m128           dummy_mask,cutoff_mask;
 461     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 462     __m128           one     = _mm_set1_ps(1.0);
 463     __m128           two     = _mm_set1_ps(2.0);
 464     x                = xx[0];
 465     f                = ff[0];
 466
 467     nri              = nlist->nri;
 468     iinr             = nlist->iinr;
 469     jindex           = nlist->jindex;
 470     jjnr             = nlist->jjnr;
 471     shiftidx         = nlist->shift;
 472     gid              = nlist->gid;
 473     shiftvec         = fr->shift_vec[0];
 474     fshift           = fr->fshift[0];
 475     nvdwtype         = fr->ntype;
 476     vdwparam         = fr->nbfp;
 477     vdwtype          = mdatoms->typeA;
 478     vdwgridparam     = fr->ljpme_c6grid;
 479     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 480     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 481     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 482
 483     rcutoff_scalar   = fr->rvdw;
 484     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 485     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 486
 487     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 488     rvdw             = _mm_set1_ps(fr->rvdw);
 489
 490     /* Avoid stupid compiler warnings */
 491     jnrA = jnrB = jnrC = jnrD = 0;
 492     j_coord_offsetA = 0;
 493     j_coord_offsetB = 0;
 494     j_coord_offsetC = 0;
 495     j_coord_offsetD = 0;
 496
 497     outeriter        = 0;
 498     inneriter        = 0;
 499
 500     for(iidx=0;iidx<4*DIM;iidx++)
 501     {
 502         scratch[iidx] = 0.0;
 503     }
 504
 505     /* Start outer loop over neighborlists */
 506     for(iidx=0; iidx<nri; iidx++)
 507     {
 508         /* Load shift vector for this list */
 509         i_shift_offset   = DIM*shiftidx[iidx];
 510
 511         /* Load limits for loop over neighbors */
 512         j_index_start    = jindex[iidx];
 513         j_index_end      = jindex[iidx+1];
 514
 515         /* Get outer coordinate index */
 516         inr              = iinr[iidx];
 517         i_coord_offset   = DIM*inr;
 518
 519         /* Load i particle coords and add shift vector */
 520         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 521
 522         fix0             = _mm_setzero_ps();
 523         fiy0             = _mm_setzero_ps();
 524         fiz0             = _mm_setzero_ps();
 525
 526         /* Load parameters for i particles */
 527         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 528
 529         /* Start inner kernel loop */
 530         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 531         {
 532
 533             /* Get j neighbor index, and coordinate index */
 534             jnrA             = jjnr[jidx];
 535             jnrB             = jjnr[jidx+1];
 536             jnrC             = jjnr[jidx+2];
 537             jnrD             = jjnr[jidx+3];
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540             j_coord_offsetC  = DIM*jnrC;
 541             j_coord_offsetD  = DIM*jnrD;
 542
 543             /* load j atom coordinates */
 544             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 545                                               x+j_coord_offsetC,x+j_coord_offsetD,
 546                                               &jx0,&jy0,&jz0);
 547
 548             /* Calculate displacement vector */
 549             dx00             = _mm_sub_ps(ix0,jx0);
 550             dy00             = _mm_sub_ps(iy0,jy0);
 551             dz00             = _mm_sub_ps(iz0,jz0);
 552
 553             /* Calculate squared distance and things based on it */
 554             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 555
 556             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 557
 558             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 559
 560             /* Load parameters for j particles */
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562             vdwjidx0B        = 2*vdwtype[jnrB+0];
 563             vdwjidx0C        = 2*vdwtype[jnrC+0];
 564             vdwjidx0D        = 2*vdwtype[jnrD+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq00,rcutoff2))
 571             {
 572
 573             r00              = _mm_mul_ps(rsq00,rinv00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 577                                          vdwparam+vdwioffset0+vdwjidx0B,
 578                                          vdwparam+vdwioffset0+vdwjidx0C,
 579                                          vdwparam+vdwioffset0+vdwjidx0D,
 580                                          &c6_00,&c12_00);
 581             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 582                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 583                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 584                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 585
 586             /* Analytical LJ-PME */
 587             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 588             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 589             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 590             exponent         = gmx_simd_exp_r(ewcljrsq);
 591             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 592             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 593             /* f6A = 6 * C6grid * (1 - poly) */
 594             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 595             /* f6B = C6grid * exponent * beta^6 */
 596             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 597             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 598             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);
 599
 600             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 601
 602             fscal            = fvdw;
 603
 604             fscal            = _mm_and_ps(fscal,cutoff_mask);
 605
 606             /* Calculate temporary vectorial force */
 607             tx               = _mm_mul_ps(fscal,dx00);
 608             ty               = _mm_mul_ps(fscal,dy00);
 609             tz               = _mm_mul_ps(fscal,dz00);
 610
 611             /* Update vectorial force */
 612             fix0             = _mm_add_ps(fix0,tx);
 613             fiy0             = _mm_add_ps(fiy0,ty);
 614             fiz0             = _mm_add_ps(fiz0,tz);
 615
 616             fjptrA             = f+j_coord_offsetA;
 617             fjptrB             = f+j_coord_offsetB;
 618             fjptrC             = f+j_coord_offsetC;
 619             fjptrD             = f+j_coord_offsetD;
 620             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 621
 622             }
 623
 624             /* Inner loop uses 49 flops */
 625         }
 626
 627         if(jidx<j_index_end)
 628         {
 629
 630             /* Get j neighbor index, and coordinate index */
 631             jnrlistA         = jjnr[jidx];
 632             jnrlistB         = jjnr[jidx+1];
 633             jnrlistC         = jjnr[jidx+2];
 634             jnrlistD         = jjnr[jidx+3];
 635             /* Sign of each element will be negative for non-real atoms.
 636              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 637              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 638              */
 639             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 640             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 641             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 642             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 643             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 644             j_coord_offsetA  = DIM*jnrA;
 645             j_coord_offsetB  = DIM*jnrB;
 646             j_coord_offsetC  = DIM*jnrC;
 647             j_coord_offsetD  = DIM*jnrD;
 648
 649             /* load j atom coordinates */
 650             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 651                                               x+j_coord_offsetC,x+j_coord_offsetD,
 652                                               &jx0,&jy0,&jz0);
 653
 654             /* Calculate displacement vector */
 655             dx00             = _mm_sub_ps(ix0,jx0);
 656             dy00             = _mm_sub_ps(iy0,jy0);
 657             dz00             = _mm_sub_ps(iz0,jz0);
 658
 659             /* Calculate squared distance and things based on it */
 660             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 661
 662             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 663
 664             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 665
 666             /* Load parameters for j particles */
 667             vdwjidx0A        = 2*vdwtype[jnrA+0];
 668             vdwjidx0B        = 2*vdwtype[jnrB+0];
 669             vdwjidx0C        = 2*vdwtype[jnrC+0];
 670             vdwjidx0D        = 2*vdwtype[jnrD+0];
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm_any_lt(rsq00,rcutoff2))
 677             {
 678
 679             r00              = _mm_mul_ps(rsq00,rinv00);
 680             r00              = _mm_andnot_ps(dummy_mask,r00);
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 684                                          vdwparam+vdwioffset0+vdwjidx0B,
 685                                          vdwparam+vdwioffset0+vdwjidx0C,
 686                                          vdwparam+vdwioffset0+vdwjidx0D,
 687                                          &c6_00,&c12_00);
 688             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 689                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 690                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 691                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 692
 693             /* Analytical LJ-PME */
 694             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 695             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 696             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 697             exponent         = gmx_simd_exp_r(ewcljrsq);
 698             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 699             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 700             /* f6A = 6 * C6grid * (1 - poly) */
 701             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 702             /* f6B = C6grid * exponent * beta^6 */
 703             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 704             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 705             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);
 706
 707             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 708
 709             fscal            = fvdw;
 710
 711             fscal            = _mm_and_ps(fscal,cutoff_mask);
 712
 713             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm_mul_ps(fscal,dx00);
 717             ty               = _mm_mul_ps(fscal,dy00);
 718             tz               = _mm_mul_ps(fscal,dz00);
 719
 720             /* Update vectorial force */
 721             fix0             = _mm_add_ps(fix0,tx);
 722             fiy0             = _mm_add_ps(fiy0,ty);
 723             fiz0             = _mm_add_ps(fiz0,tz);
 724
 725             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 726             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 727             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 728             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 729             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 730
 731             }
 732
 733             /* Inner loop uses 50 flops */
 734         }
 735
 736         /* End of innermost loop */
 737
 738         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 739                                               f+i_coord_offset,fshift+i_shift_offset);
 740
 741         /* Increment number of inner iterations */
 742         inneriter                  += j_index_end - j_index_start;
 743
 744         /* Outer loop uses 6 flops */
 745     }
 746
 747     /* Increment number of outer iterations */
 748     outeriter        += nri;
 749
 750     /* Update outer/inner flops */
 751
 752     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);
 753 }