src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  95     real             rswitch_scalar,d_scalar;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     rcutoff_scalar   = fr->rvdw;
 116     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 117     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 118
 119     rswitch_scalar   = fr->rvdw_switch;
 120     rswitch          = _mm_set1_ps(rswitch_scalar);
 121     /* Setup switch parameters */
 122     d_scalar         = rcutoff_scalar-rswitch_scalar;
 123     d                = _mm_set1_ps(d_scalar);
 124     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 125     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 126     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 127     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 128     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 129     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166
 167         /* Load parameters for i particles */
 168         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 169
 170         /* Reset potential sums */
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 201
 202             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             vdwjidx0A        = 2*vdwtype[jnrA+0];
 206             vdwjidx0B        = 2*vdwtype[jnrB+0];
 207             vdwjidx0C        = 2*vdwtype[jnrC+0];
 208             vdwjidx0D        = 2*vdwtype[jnrD+0];
 209
 210             /**************************
 211              * CALCULATE INTERACTIONS *
 212              **************************/
 213
 214             if (gmx_mm_any_lt(rsq00,rcutoff2))
 215             {
 216
 217             r00              = _mm_mul_ps(rsq00,rinv00);
 218
 219             /* Compute parameters for interactions between i and j atoms */
 220             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,
 222                                          vdwparam+vdwioffset0+vdwjidx0C,
 223                                          vdwparam+vdwioffset0+vdwjidx0D,
 224                                          &c6_00,&c12_00);
 225
 226             /* LENNARD-JONES DISPERSION/REPULSION */
 227
 228             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 229             vvdw6            = _mm_mul_ps(c6_00,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             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 233
 234             d                = _mm_sub_ps(r00,rswitch);
 235             d                = _mm_max_ps(d,_mm_setzero_ps());
 236             d2               = _mm_mul_ps(d,d);
 237             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 238
 239             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 240
 241             /* Evaluate switch function */
 242             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 243             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 244             vvdw             = _mm_mul_ps(vvdw,sw);
 245             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 249             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 250
 251             fscal            = fvdw;
 252
 253             fscal            = _mm_and_ps(fscal,cutoff_mask);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_ps(fscal,dx00);
 257             ty               = _mm_mul_ps(fscal,dy00);
 258             tz               = _mm_mul_ps(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_ps(fix0,tx);
 262             fiy0             = _mm_add_ps(fiy0,ty);
 263             fiz0             = _mm_add_ps(fiz0,tz);
 264
 265             fjptrA             = f+j_coord_offsetA;
 266             fjptrB             = f+j_coord_offsetB;
 267             fjptrC             = f+j_coord_offsetC;
 268             fjptrD             = f+j_coord_offsetD;
 269             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 270
 271             }
 272
 273             /* Inner loop uses 59 flops */
 274         }
 275
 276         if(jidx<j_index_end)
 277         {
 278
 279             /* Get j neighbor index, and coordinate index */
 280             jnrlistA         = jjnr[jidx];
 281             jnrlistB         = jjnr[jidx+1];
 282             jnrlistC         = jjnr[jidx+2];
 283             jnrlistD         = jjnr[jidx+3];
 284             /* Sign of each element will be negative for non-real atoms.
 285              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 286              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 287              */
 288             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 289             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 290             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 291             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 292             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 293             j_coord_offsetA  = DIM*jnrA;
 294             j_coord_offsetB  = DIM*jnrB;
 295             j_coord_offsetC  = DIM*jnrC;
 296             j_coord_offsetD  = DIM*jnrD;
 297
 298             /* load j atom coordinates */
 299             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 300                                               x+j_coord_offsetC,x+j_coord_offsetD,
 301                                               &jx0,&jy0,&jz0);
 302
 303             /* Calculate displacement vector */
 304             dx00             = _mm_sub_ps(ix0,jx0);
 305             dy00             = _mm_sub_ps(iy0,jy0);
 306             dz00             = _mm_sub_ps(iz0,jz0);
 307
 308             /* Calculate squared distance and things based on it */
 309             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 310
 311             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 312
 313             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 314
 315             /* Load parameters for j particles */
 316             vdwjidx0A        = 2*vdwtype[jnrA+0];
 317             vdwjidx0B        = 2*vdwtype[jnrB+0];
 318             vdwjidx0C        = 2*vdwtype[jnrC+0];
 319             vdwjidx0D        = 2*vdwtype[jnrD+0];
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq00,rcutoff2))
 326             {
 327
 328             r00              = _mm_mul_ps(rsq00,rinv00);
 329             r00              = _mm_andnot_ps(dummy_mask,r00);
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 333                                          vdwparam+vdwioffset0+vdwjidx0B,
 334                                          vdwparam+vdwioffset0+vdwjidx0C,
 335                                          vdwparam+vdwioffset0+vdwjidx0D,
 336                                          &c6_00,&c12_00);
 337
 338             /* LENNARD-JONES DISPERSION/REPULSION */
 339
 340             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 341             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 342             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 343             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 344             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 345
 346             d                = _mm_sub_ps(r00,rswitch);
 347             d                = _mm_max_ps(d,_mm_setzero_ps());
 348             d2               = _mm_mul_ps(d,d);
 349             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 350
 351             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 352
 353             /* Evaluate switch function */
 354             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 355             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 356             vvdw             = _mm_mul_ps(vvdw,sw);
 357             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 361             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 362             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 363
 364             fscal            = fvdw;
 365
 366             fscal            = _mm_and_ps(fscal,cutoff_mask);
 367
 368             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm_mul_ps(fscal,dx00);
 372             ty               = _mm_mul_ps(fscal,dy00);
 373             tz               = _mm_mul_ps(fscal,dz00);
 374
 375             /* Update vectorial force */
 376             fix0             = _mm_add_ps(fix0,tx);
 377             fiy0             = _mm_add_ps(fiy0,ty);
 378             fiz0             = _mm_add_ps(fiz0,tz);
 379
 380             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 381             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 382             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 383             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 384             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 385
 386             }
 387
 388             /* Inner loop uses 60 flops */
 389         }
 390
 391         /* End of innermost loop */
 392
 393         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 394                                               f+i_coord_offset,fshift+i_shift_offset);
 395
 396         ggid                        = gid[iidx];
 397         /* Update potential energies */
 398         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 399
 400         /* Increment number of inner iterations */
 401         inneriter                  += j_index_end - j_index_start;
 402
 403         /* Outer loop uses 7 flops */
 404     }
 405
 406     /* Increment number of outer iterations */
 407     outeriter        += nri;
 408
 409     /* Update outer/inner flops */
 410
 411     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
 412 }
 413 /*
 414  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 415  * Electrostatics interaction: None
 416  * VdW interaction:            LennardJones
 417  * Geometry:                   Particle-Particle
 418  * Calculate force/pot:        Force
 419  */
 420 void
 421 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 422                     (t_nblist                    * gmx_restrict       nlist,
 423                      rvec                        * gmx_restrict          xx,
 424                      rvec                        * gmx_restrict          ff,
 425                      t_forcerec                  * gmx_restrict          fr,
 426                      t_mdatoms                   * gmx_restrict     mdatoms,
 427                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 428                      t_nrnb                      * gmx_restrict        nrnb)
 429 {
 430     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 431      * just 0 for non-waters.
 432      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 433      * jnr indices corresponding to data put in the four positions in the SIMD register.
 434      */
 435     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 436     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 437     int              jnrA,jnrB,jnrC,jnrD;
 438     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 439     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 440     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 441     real             rcutoff_scalar;
 442     real             *shiftvec,*fshift,*x,*f;
 443     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 444     real             scratch[4*DIM];
 445     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 446     int              vdwioffset0;
 447     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 448     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 449     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 450     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 451     int              nvdwtype;
 452     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 453     int              *vdwtype;
 454     real             *vdwparam;
 455     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 456     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 457     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 458     real             rswitch_scalar,d_scalar;
 459     __m128           dummy_mask,cutoff_mask;
 460     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 461     __m128           one     = _mm_set1_ps(1.0);
 462     __m128           two     = _mm_set1_ps(2.0);
 463     x                = xx[0];
 464     f                = ff[0];
 465
 466     nri              = nlist->nri;
 467     iinr             = nlist->iinr;
 468     jindex           = nlist->jindex;
 469     jjnr             = nlist->jjnr;
 470     shiftidx         = nlist->shift;
 471     gid              = nlist->gid;
 472     shiftvec         = fr->shift_vec[0];
 473     fshift           = fr->fshift[0];
 474     nvdwtype         = fr->ntype;
 475     vdwparam         = fr->nbfp;
 476     vdwtype          = mdatoms->typeA;
 477
 478     rcutoff_scalar   = fr->rvdw;
 479     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 480     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 481
 482     rswitch_scalar   = fr->rvdw_switch;
 483     rswitch          = _mm_set1_ps(rswitch_scalar);
 484     /* Setup switch parameters */
 485     d_scalar         = rcutoff_scalar-rswitch_scalar;
 486     d                = _mm_set1_ps(d_scalar);
 487     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 488     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 489     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 490     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 491     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 492     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 493
 494     /* Avoid stupid compiler warnings */
 495     jnrA = jnrB = jnrC = jnrD = 0;
 496     j_coord_offsetA = 0;
 497     j_coord_offsetB = 0;
 498     j_coord_offsetC = 0;
 499     j_coord_offsetD = 0;
 500
 501     outeriter        = 0;
 502     inneriter        = 0;
 503
 504     for(iidx=0;iidx<4*DIM;iidx++)
 505     {
 506         scratch[iidx] = 0.0;
 507     }
 508
 509     /* Start outer loop over neighborlists */
 510     for(iidx=0; iidx<nri; iidx++)
 511     {
 512         /* Load shift vector for this list */
 513         i_shift_offset   = DIM*shiftidx[iidx];
 514
 515         /* Load limits for loop over neighbors */
 516         j_index_start    = jindex[iidx];
 517         j_index_end      = jindex[iidx+1];
 518
 519         /* Get outer coordinate index */
 520         inr              = iinr[iidx];
 521         i_coord_offset   = DIM*inr;
 522
 523         /* Load i particle coords and add shift vector */
 524         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 525
 526         fix0             = _mm_setzero_ps();
 527         fiy0             = _mm_setzero_ps();
 528         fiz0             = _mm_setzero_ps();
 529
 530         /* Load parameters for i particles */
 531         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 532
 533         /* Start inner kernel loop */
 534         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrA             = jjnr[jidx];
 539             jnrB             = jjnr[jidx+1];
 540             jnrC             = jjnr[jidx+2];
 541             jnrD             = jjnr[jidx+3];
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544             j_coord_offsetC  = DIM*jnrC;
 545             j_coord_offsetD  = DIM*jnrD;
 546
 547             /* load j atom coordinates */
 548             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                               x+j_coord_offsetC,x+j_coord_offsetD,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm_sub_ps(ix0,jx0);
 554             dy00             = _mm_sub_ps(iy0,jy0);
 555             dz00             = _mm_sub_ps(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 561
 562             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             vdwjidx0A        = 2*vdwtype[jnrA+0];
 566             vdwjidx0B        = 2*vdwtype[jnrB+0];
 567             vdwjidx0C        = 2*vdwtype[jnrC+0];
 568             vdwjidx0D        = 2*vdwtype[jnrD+0];
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             if (gmx_mm_any_lt(rsq00,rcutoff2))
 575             {
 576
 577             r00              = _mm_mul_ps(rsq00,rinv00);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 581                                          vdwparam+vdwioffset0+vdwjidx0B,
 582                                          vdwparam+vdwioffset0+vdwjidx0C,
 583                                          vdwparam+vdwioffset0+vdwjidx0D,
 584                                          &c6_00,&c12_00);
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 589             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 590             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 591             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 592             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 593
 594             d                = _mm_sub_ps(r00,rswitch);
 595             d                = _mm_max_ps(d,_mm_setzero_ps());
 596             d2               = _mm_mul_ps(d,d);
 597             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 598
 599             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 600
 601             /* Evaluate switch function */
 602             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 603             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 604             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 605
 606             fscal            = fvdw;
 607
 608             fscal            = _mm_and_ps(fscal,cutoff_mask);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm_mul_ps(fscal,dx00);
 612             ty               = _mm_mul_ps(fscal,dy00);
 613             tz               = _mm_mul_ps(fscal,dz00);
 614
 615             /* Update vectorial force */
 616             fix0             = _mm_add_ps(fix0,tx);
 617             fiy0             = _mm_add_ps(fiy0,ty);
 618             fiz0             = _mm_add_ps(fiz0,tz);
 619
 620             fjptrA             = f+j_coord_offsetA;
 621             fjptrB             = f+j_coord_offsetB;
 622             fjptrC             = f+j_coord_offsetC;
 623             fjptrD             = f+j_coord_offsetD;
 624             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 625
 626             }
 627
 628             /* Inner loop uses 56 flops */
 629         }
 630
 631         if(jidx<j_index_end)
 632         {
 633
 634             /* Get j neighbor index, and coordinate index */
 635             jnrlistA         = jjnr[jidx];
 636             jnrlistB         = jjnr[jidx+1];
 637             jnrlistC         = jjnr[jidx+2];
 638             jnrlistD         = jjnr[jidx+3];
 639             /* Sign of each element will be negative for non-real atoms.
 640              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 641              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 642              */
 643             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 644             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 645             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 646             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 647             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 648             j_coord_offsetA  = DIM*jnrA;
 649             j_coord_offsetB  = DIM*jnrB;
 650             j_coord_offsetC  = DIM*jnrC;
 651             j_coord_offsetD  = DIM*jnrD;
 652
 653             /* load j atom coordinates */
 654             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 655                                               x+j_coord_offsetC,x+j_coord_offsetD,
 656                                               &jx0,&jy0,&jz0);
 657
 658             /* Calculate displacement vector */
 659             dx00             = _mm_sub_ps(ix0,jx0);
 660             dy00             = _mm_sub_ps(iy0,jy0);
 661             dz00             = _mm_sub_ps(iz0,jz0);
 662
 663             /* Calculate squared distance and things based on it */
 664             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 665
 666             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 667
 668             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 669
 670             /* Load parameters for j particles */
 671             vdwjidx0A        = 2*vdwtype[jnrA+0];
 672             vdwjidx0B        = 2*vdwtype[jnrB+0];
 673             vdwjidx0C        = 2*vdwtype[jnrC+0];
 674             vdwjidx0D        = 2*vdwtype[jnrD+0];
 675
 676             /**************************
 677              * CALCULATE INTERACTIONS *
 678              **************************/
 679
 680             if (gmx_mm_any_lt(rsq00,rcutoff2))
 681             {
 682
 683             r00              = _mm_mul_ps(rsq00,rinv00);
 684             r00              = _mm_andnot_ps(dummy_mask,r00);
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 688                                          vdwparam+vdwioffset0+vdwjidx0B,
 689                                          vdwparam+vdwioffset0+vdwjidx0C,
 690                                          vdwparam+vdwioffset0+vdwjidx0D,
 691                                          &c6_00,&c12_00);
 692
 693             /* LENNARD-JONES DISPERSION/REPULSION */
 694
 695             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 696             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 697             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 698             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 699             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 700
 701             d                = _mm_sub_ps(r00,rswitch);
 702             d                = _mm_max_ps(d,_mm_setzero_ps());
 703             d2               = _mm_mul_ps(d,d);
 704             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 705
 706             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 707
 708             /* Evaluate switch function */
 709             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 710             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 711             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 712
 713             fscal            = fvdw;
 714
 715             fscal            = _mm_and_ps(fscal,cutoff_mask);
 716
 717             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm_mul_ps(fscal,dx00);
 721             ty               = _mm_mul_ps(fscal,dy00);
 722             tz               = _mm_mul_ps(fscal,dz00);
 723
 724             /* Update vectorial force */
 725             fix0             = _mm_add_ps(fix0,tx);
 726             fiy0             = _mm_add_ps(fiy0,ty);
 727             fiz0             = _mm_add_ps(fiz0,tz);
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 734
 735             }
 736
 737             /* Inner loop uses 57 flops */
 738         }
 739
 740         /* End of innermost loop */
 741
 742         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 743                                               f+i_coord_offset,fshift+i_shift_offset);
 744
 745         /* Increment number of inner iterations */
 746         inneriter                  += j_index_end - j_index_start;
 747
 748         /* Outer loop uses 6 flops */
 749     }
 750
 751     /* Increment number of outer iterations */
 752     outeriter        += nri;
 753
 754     /* Update outer/inner flops */
 755
 756     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);
 757 }