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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  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
 117     rcutoff_scalar   = fr->rvdw;
 118     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 119     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 120
 121     rswitch_scalar   = fr->rvdw_switch;
 122     rswitch          = _mm_set1_ps(rswitch_scalar);
 123     /* Setup switch parameters */
 124     d_scalar         = rcutoff_scalar-rswitch_scalar;
 125     d                = _mm_set1_ps(d_scalar);
 126     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 127     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 128     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 129     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 130     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 131     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         vvdwsum          = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_ps(ix0,jx0);
 196             dy00             = _mm_sub_ps(iy0,jy0);
 197             dz00             = _mm_sub_ps(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 203
 204             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209             vdwjidx0C        = 2*vdwtype[jnrC+0];
 210             vdwjidx0D        = 2*vdwtype[jnrD+0];
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             if (gmx_mm_any_lt(rsq00,rcutoff2))
 217             {
 218
 219             r00              = _mm_mul_ps(rsq00,rinv00);
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,
 224                                          vdwparam+vdwioffset0+vdwjidx0C,
 225                                          vdwparam+vdwioffset0+vdwjidx0D,
 226                                          &c6_00,&c12_00);
 227
 228             /* LENNARD-JONES DISPERSION/REPULSION */
 229
 230             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 231             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 232             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 233             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 234             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 235
 236             d                = _mm_sub_ps(r00,rswitch);
 237             d                = _mm_max_ps(d,_mm_setzero_ps());
 238             d2               = _mm_mul_ps(d,d);
 239             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)))))));
 240
 241             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 242
 243             /* Evaluate switch function */
 244             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 245             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 246             vvdw             = _mm_mul_ps(vvdw,sw);
 247             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 251             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 252
 253             fscal            = fvdw;
 254
 255             fscal            = _mm_and_ps(fscal,cutoff_mask);
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm_mul_ps(fscal,dx00);
 259             ty               = _mm_mul_ps(fscal,dy00);
 260             tz               = _mm_mul_ps(fscal,dz00);
 261
 262             /* Update vectorial force */
 263             fix0             = _mm_add_ps(fix0,tx);
 264             fiy0             = _mm_add_ps(fiy0,ty);
 265             fiz0             = _mm_add_ps(fiz0,tz);
 266
 267             fjptrA             = f+j_coord_offsetA;
 268             fjptrB             = f+j_coord_offsetB;
 269             fjptrC             = f+j_coord_offsetC;
 270             fjptrD             = f+j_coord_offsetD;
 271             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 272
 273             }
 274
 275             /* Inner loop uses 59 flops */
 276         }
 277
 278         if(jidx<j_index_end)
 279         {
 280
 281             /* Get j neighbor index, and coordinate index */
 282             jnrlistA         = jjnr[jidx];
 283             jnrlistB         = jjnr[jidx+1];
 284             jnrlistC         = jjnr[jidx+2];
 285             jnrlistD         = jjnr[jidx+3];
 286             /* Sign of each element will be negative for non-real atoms.
 287              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 288              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 289              */
 290             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 291             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 292             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 293             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 294             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 295             j_coord_offsetA  = DIM*jnrA;
 296             j_coord_offsetB  = DIM*jnrB;
 297             j_coord_offsetC  = DIM*jnrC;
 298             j_coord_offsetD  = DIM*jnrD;
 299
 300             /* load j atom coordinates */
 301             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 302                                               x+j_coord_offsetC,x+j_coord_offsetD,
 303                                               &jx0,&jy0,&jz0);
 304
 305             /* Calculate displacement vector */
 306             dx00             = _mm_sub_ps(ix0,jx0);
 307             dy00             = _mm_sub_ps(iy0,jy0);
 308             dz00             = _mm_sub_ps(iz0,jz0);
 309
 310             /* Calculate squared distance and things based on it */
 311             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 312
 313             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 314
 315             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 316
 317             /* Load parameters for j particles */
 318             vdwjidx0A        = 2*vdwtype[jnrA+0];
 319             vdwjidx0B        = 2*vdwtype[jnrB+0];
 320             vdwjidx0C        = 2*vdwtype[jnrC+0];
 321             vdwjidx0D        = 2*vdwtype[jnrD+0];
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             if (gmx_mm_any_lt(rsq00,rcutoff2))
 328             {
 329
 330             r00              = _mm_mul_ps(rsq00,rinv00);
 331             r00              = _mm_andnot_ps(dummy_mask,r00);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 335                                          vdwparam+vdwioffset0+vdwjidx0B,
 336                                          vdwparam+vdwioffset0+vdwjidx0C,
 337                                          vdwparam+vdwioffset0+vdwjidx0D,
 338                                          &c6_00,&c12_00);
 339
 340             /* LENNARD-JONES DISPERSION/REPULSION */
 341
 342             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 343             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 344             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 345             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 346             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 347
 348             d                = _mm_sub_ps(r00,rswitch);
 349             d                = _mm_max_ps(d,_mm_setzero_ps());
 350             d2               = _mm_mul_ps(d,d);
 351             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)))))));
 352
 353             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 354
 355             /* Evaluate switch function */
 356             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 357             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 358             vvdw             = _mm_mul_ps(vvdw,sw);
 359             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 363             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 364             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 365
 366             fscal            = fvdw;
 367
 368             fscal            = _mm_and_ps(fscal,cutoff_mask);
 369
 370             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_ps(fscal,dx00);
 374             ty               = _mm_mul_ps(fscal,dy00);
 375             tz               = _mm_mul_ps(fscal,dz00);
 376
 377             /* Update vectorial force */
 378             fix0             = _mm_add_ps(fix0,tx);
 379             fiy0             = _mm_add_ps(fiy0,ty);
 380             fiz0             = _mm_add_ps(fiz0,tz);
 381
 382             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 383             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 384             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 385             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 387
 388             }
 389
 390             /* Inner loop uses 60 flops */
 391         }
 392
 393         /* End of innermost loop */
 394
 395         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 396                                               f+i_coord_offset,fshift+i_shift_offset);
 397
 398         ggid                        = gid[iidx];
 399         /* Update potential energies */
 400         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 401
 402         /* Increment number of inner iterations */
 403         inneriter                  += j_index_end - j_index_start;
 404
 405         /* Outer loop uses 7 flops */
 406     }
 407
 408     /* Increment number of outer iterations */
 409     outeriter        += nri;
 410
 411     /* Update outer/inner flops */
 412
 413     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
 414 }
 415 /*
 416  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 417  * Electrostatics interaction: None
 418  * VdW interaction:            LennardJones
 419  * Geometry:                   Particle-Particle
 420  * Calculate force/pot:        Force
 421  */
 422 void
 423 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 424                     (t_nblist                    * gmx_restrict       nlist,
 425                      rvec                        * gmx_restrict          xx,
 426                      rvec                        * gmx_restrict          ff,
 427                      t_forcerec                  * gmx_restrict          fr,
 428                      t_mdatoms                   * gmx_restrict     mdatoms,
 429                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 430                      t_nrnb                      * gmx_restrict        nrnb)
 431 {
 432     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 433      * just 0 for non-waters.
 434      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 435      * jnr indices corresponding to data put in the four positions in the SIMD register.
 436      */
 437     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 438     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 439     int              jnrA,jnrB,jnrC,jnrD;
 440     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 441     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 442     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 443     real             rcutoff_scalar;
 444     real             *shiftvec,*fshift,*x,*f;
 445     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 446     real             scratch[4*DIM];
 447     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 448     int              vdwioffset0;
 449     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 450     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 451     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 452     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 453     int              nvdwtype;
 454     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 455     int              *vdwtype;
 456     real             *vdwparam;
 457     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 458     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 459     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 460     real             rswitch_scalar,d_scalar;
 461     __m128           dummy_mask,cutoff_mask;
 462     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 463     __m128           one     = _mm_set1_ps(1.0);
 464     __m128           two     = _mm_set1_ps(2.0);
 465     x                = xx[0];
 466     f                = ff[0];
 467
 468     nri              = nlist->nri;
 469     iinr             = nlist->iinr;
 470     jindex           = nlist->jindex;
 471     jjnr             = nlist->jjnr;
 472     shiftidx         = nlist->shift;
 473     gid              = nlist->gid;
 474     shiftvec         = fr->shift_vec[0];
 475     fshift           = fr->fshift[0];
 476     nvdwtype         = fr->ntype;
 477     vdwparam         = fr->nbfp;
 478     vdwtype          = mdatoms->typeA;
 479
 480     rcutoff_scalar   = fr->rvdw;
 481     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 482     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 483
 484     rswitch_scalar   = fr->rvdw_switch;
 485     rswitch          = _mm_set1_ps(rswitch_scalar);
 486     /* Setup switch parameters */
 487     d_scalar         = rcutoff_scalar-rswitch_scalar;
 488     d                = _mm_set1_ps(d_scalar);
 489     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 490     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 491     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 492     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 493     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 494     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 495
 496     /* Avoid stupid compiler warnings */
 497     jnrA = jnrB = jnrC = jnrD = 0;
 498     j_coord_offsetA = 0;
 499     j_coord_offsetB = 0;
 500     j_coord_offsetC = 0;
 501     j_coord_offsetD = 0;
 502
 503     outeriter        = 0;
 504     inneriter        = 0;
 505
 506     for(iidx=0;iidx<4*DIM;iidx++)
 507     {
 508         scratch[iidx] = 0.0;
 509     }
 510
 511     /* Start outer loop over neighborlists */
 512     for(iidx=0; iidx<nri; iidx++)
 513     {
 514         /* Load shift vector for this list */
 515         i_shift_offset   = DIM*shiftidx[iidx];
 516
 517         /* Load limits for loop over neighbors */
 518         j_index_start    = jindex[iidx];
 519         j_index_end      = jindex[iidx+1];
 520
 521         /* Get outer coordinate index */
 522         inr              = iinr[iidx];
 523         i_coord_offset   = DIM*inr;
 524
 525         /* Load i particle coords and add shift vector */
 526         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 527
 528         fix0             = _mm_setzero_ps();
 529         fiy0             = _mm_setzero_ps();
 530         fiz0             = _mm_setzero_ps();
 531
 532         /* Load parameters for i particles */
 533         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 534
 535         /* Start inner kernel loop */
 536         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 537         {
 538
 539             /* Get j neighbor index, and coordinate index */
 540             jnrA             = jjnr[jidx];
 541             jnrB             = jjnr[jidx+1];
 542             jnrC             = jjnr[jidx+2];
 543             jnrD             = jjnr[jidx+3];
 544             j_coord_offsetA  = DIM*jnrA;
 545             j_coord_offsetB  = DIM*jnrB;
 546             j_coord_offsetC  = DIM*jnrC;
 547             j_coord_offsetD  = DIM*jnrD;
 548
 549             /* load j atom coordinates */
 550             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 551                                               x+j_coord_offsetC,x+j_coord_offsetD,
 552                                               &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm_sub_ps(ix0,jx0);
 556             dy00             = _mm_sub_ps(iy0,jy0);
 557             dz00             = _mm_sub_ps(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 563
 564             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568             vdwjidx0B        = 2*vdwtype[jnrB+0];
 569             vdwjidx0C        = 2*vdwtype[jnrC+0];
 570             vdwjidx0D        = 2*vdwtype[jnrD+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_mm_any_lt(rsq00,rcutoff2))
 577             {
 578
 579             r00              = _mm_mul_ps(rsq00,rinv00);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 583                                          vdwparam+vdwioffset0+vdwjidx0B,
 584                                          vdwparam+vdwioffset0+vdwjidx0C,
 585                                          vdwparam+vdwioffset0+vdwjidx0D,
 586                                          &c6_00,&c12_00);
 587
 588             /* LENNARD-JONES DISPERSION/REPULSION */
 589
 590             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 591             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 592             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 593             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 594             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 595
 596             d                = _mm_sub_ps(r00,rswitch);
 597             d                = _mm_max_ps(d,_mm_setzero_ps());
 598             d2               = _mm_mul_ps(d,d);
 599             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)))))));
 600
 601             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 602
 603             /* Evaluate switch function */
 604             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 605             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 606             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 607
 608             fscal            = fvdw;
 609
 610             fscal            = _mm_and_ps(fscal,cutoff_mask);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm_mul_ps(fscal,dx00);
 614             ty               = _mm_mul_ps(fscal,dy00);
 615             tz               = _mm_mul_ps(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm_add_ps(fix0,tx);
 619             fiy0             = _mm_add_ps(fiy0,ty);
 620             fiz0             = _mm_add_ps(fiz0,tz);
 621
 622             fjptrA             = f+j_coord_offsetA;
 623             fjptrB             = f+j_coord_offsetB;
 624             fjptrC             = f+j_coord_offsetC;
 625             fjptrD             = f+j_coord_offsetD;
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 627
 628             }
 629
 630             /* Inner loop uses 56 flops */
 631         }
 632
 633         if(jidx<j_index_end)
 634         {
 635
 636             /* Get j neighbor index, and coordinate index */
 637             jnrlistA         = jjnr[jidx];
 638             jnrlistB         = jjnr[jidx+1];
 639             jnrlistC         = jjnr[jidx+2];
 640             jnrlistD         = jjnr[jidx+3];
 641             /* Sign of each element will be negative for non-real atoms.
 642              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 643              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 644              */
 645             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 646             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 647             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 648             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 649             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 650             j_coord_offsetA  = DIM*jnrA;
 651             j_coord_offsetB  = DIM*jnrB;
 652             j_coord_offsetC  = DIM*jnrC;
 653             j_coord_offsetD  = DIM*jnrD;
 654
 655             /* load j atom coordinates */
 656             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 657                                               x+j_coord_offsetC,x+j_coord_offsetD,
 658                                               &jx0,&jy0,&jz0);
 659
 660             /* Calculate displacement vector */
 661             dx00             = _mm_sub_ps(ix0,jx0);
 662             dy00             = _mm_sub_ps(iy0,jy0);
 663             dz00             = _mm_sub_ps(iz0,jz0);
 664
 665             /* Calculate squared distance and things based on it */
 666             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 667
 668             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 669
 670             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 671
 672             /* Load parameters for j particles */
 673             vdwjidx0A        = 2*vdwtype[jnrA+0];
 674             vdwjidx0B        = 2*vdwtype[jnrB+0];
 675             vdwjidx0C        = 2*vdwtype[jnrC+0];
 676             vdwjidx0D        = 2*vdwtype[jnrD+0];
 677
 678             /**************************
 679              * CALCULATE INTERACTIONS *
 680              **************************/
 681
 682             if (gmx_mm_any_lt(rsq00,rcutoff2))
 683             {
 684
 685             r00              = _mm_mul_ps(rsq00,rinv00);
 686             r00              = _mm_andnot_ps(dummy_mask,r00);
 687
 688             /* Compute parameters for interactions between i and j atoms */
 689             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 690                                          vdwparam+vdwioffset0+vdwjidx0B,
 691                                          vdwparam+vdwioffset0+vdwjidx0C,
 692                                          vdwparam+vdwioffset0+vdwjidx0D,
 693                                          &c6_00,&c12_00);
 694
 695             /* LENNARD-JONES DISPERSION/REPULSION */
 696
 697             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 698             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 699             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 700             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 701             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 702
 703             d                = _mm_sub_ps(r00,rswitch);
 704             d                = _mm_max_ps(d,_mm_setzero_ps());
 705             d2               = _mm_mul_ps(d,d);
 706             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)))))));
 707
 708             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 709
 710             /* Evaluate switch function */
 711             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 712             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 713             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 714
 715             fscal            = fvdw;
 716
 717             fscal            = _mm_and_ps(fscal,cutoff_mask);
 718
 719             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm_mul_ps(fscal,dx00);
 723             ty               = _mm_mul_ps(fscal,dy00);
 724             tz               = _mm_mul_ps(fscal,dz00);
 725
 726             /* Update vectorial force */
 727             fix0             = _mm_add_ps(fix0,tx);
 728             fiy0             = _mm_add_ps(fiy0,ty);
 729             fiz0             = _mm_add_ps(fiz0,tz);
 730
 731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 735             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 736
 737             }
 738
 739             /* Inner loop uses 57 flops */
 740         }
 741
 742         /* End of innermost loop */
 743
 744         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 745                                               f+i_coord_offset,fshift+i_shift_offset);
 746
 747         /* Increment number of inner iterations */
 748         inneriter                  += j_index_end - j_index_start;
 749
 750         /* Outer loop uses 6 flops */
 751     }
 752
 753     /* Increment number of outer iterations */
 754     outeriter        += nri;
 755
 756     /* Update outer/inner flops */
 757
 758     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);
 759 }