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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  92     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  93     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  94     real             rswitch_scalar,d_scalar;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     rcutoff_scalar   = fr->ic->rvdw;
 115     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 117
 118     rswitch_scalar   = fr->ic->rvdw_switch;
 119     rswitch          = _mm_set1_ps(rswitch_scalar);
 120     /* Setup switch parameters */
 121     d_scalar         = rcutoff_scalar-rswitch_scalar;
 122     d                = _mm_set1_ps(d_scalar);
 123     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 124     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 127     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 128     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165
 166         /* Load parameters for i particles */
 167         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 168
 169         /* Reset potential sums */
 170         vvdwsum          = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               x+j_coord_offsetC,x+j_coord_offsetD,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_ps(ix0,jx0);
 193             dy00             = _mm_sub_ps(iy0,jy0);
 194             dz00             = _mm_sub_ps(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 198
 199             rinv00           = sse2_invsqrt_f(rsq00);
 200
 201             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 202
 203             /* Load parameters for j particles */
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206             vdwjidx0C        = 2*vdwtype[jnrC+0];
 207             vdwjidx0D        = 2*vdwtype[jnrD+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_mm_any_lt(rsq00,rcutoff2))
 214             {
 215
 216             r00              = _mm_mul_ps(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 220                                          vdwparam+vdwioffset0+vdwjidx0B,
 221                                          vdwparam+vdwioffset0+vdwjidx0C,
 222                                          vdwparam+vdwioffset0+vdwjidx0D,
 223                                          &c6_00,&c12_00);
 224
 225             /* LENNARD-JONES DISPERSION/REPULSION */
 226
 227             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 228             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 229             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 230             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 231             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 232
 233             d                = _mm_sub_ps(r00,rswitch);
 234             d                = _mm_max_ps(d,_mm_setzero_ps());
 235             d2               = _mm_mul_ps(d,d);
 236             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)))))));
 237
 238             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 239
 240             /* Evaluate switch function */
 241             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 242             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 243             vvdw             = _mm_mul_ps(vvdw,sw);
 244             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 248             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 249
 250             fscal            = fvdw;
 251
 252             fscal            = _mm_and_ps(fscal,cutoff_mask);
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm_mul_ps(fscal,dx00);
 256             ty               = _mm_mul_ps(fscal,dy00);
 257             tz               = _mm_mul_ps(fscal,dz00);
 258
 259             /* Update vectorial force */
 260             fix0             = _mm_add_ps(fix0,tx);
 261             fiy0             = _mm_add_ps(fiy0,ty);
 262             fiz0             = _mm_add_ps(fiz0,tz);
 263
 264             fjptrA             = f+j_coord_offsetA;
 265             fjptrB             = f+j_coord_offsetB;
 266             fjptrC             = f+j_coord_offsetC;
 267             fjptrD             = f+j_coord_offsetD;
 268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 269
 270             }
 271
 272             /* Inner loop uses 59 flops */
 273         }
 274
 275         if(jidx<j_index_end)
 276         {
 277
 278             /* Get j neighbor index, and coordinate index */
 279             jnrlistA         = jjnr[jidx];
 280             jnrlistB         = jjnr[jidx+1];
 281             jnrlistC         = jjnr[jidx+2];
 282             jnrlistD         = jjnr[jidx+3];
 283             /* Sign of each element will be negative for non-real atoms.
 284              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 285              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 286              */
 287             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 288             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 289             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 290             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 291             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 292             j_coord_offsetA  = DIM*jnrA;
 293             j_coord_offsetB  = DIM*jnrB;
 294             j_coord_offsetC  = DIM*jnrC;
 295             j_coord_offsetD  = DIM*jnrD;
 296
 297             /* load j atom coordinates */
 298             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 299                                               x+j_coord_offsetC,x+j_coord_offsetD,
 300                                               &jx0,&jy0,&jz0);
 301
 302             /* Calculate displacement vector */
 303             dx00             = _mm_sub_ps(ix0,jx0);
 304             dy00             = _mm_sub_ps(iy0,jy0);
 305             dz00             = _mm_sub_ps(iz0,jz0);
 306
 307             /* Calculate squared distance and things based on it */
 308             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 309
 310             rinv00           = sse2_invsqrt_f(rsq00);
 311
 312             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 313
 314             /* Load parameters for j particles */
 315             vdwjidx0A        = 2*vdwtype[jnrA+0];
 316             vdwjidx0B        = 2*vdwtype[jnrB+0];
 317             vdwjidx0C        = 2*vdwtype[jnrC+0];
 318             vdwjidx0D        = 2*vdwtype[jnrD+0];
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             if (gmx_mm_any_lt(rsq00,rcutoff2))
 325             {
 326
 327             r00              = _mm_mul_ps(rsq00,rinv00);
 328             r00              = _mm_andnot_ps(dummy_mask,r00);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 332                                          vdwparam+vdwioffset0+vdwjidx0B,
 333                                          vdwparam+vdwioffset0+vdwjidx0C,
 334                                          vdwparam+vdwioffset0+vdwjidx0D,
 335                                          &c6_00,&c12_00);
 336
 337             /* LENNARD-JONES DISPERSION/REPULSION */
 338
 339             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 340             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 341             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 342             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 343             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 344
 345             d                = _mm_sub_ps(r00,rswitch);
 346             d                = _mm_max_ps(d,_mm_setzero_ps());
 347             d2               = _mm_mul_ps(d,d);
 348             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)))))));
 349
 350             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 351
 352             /* Evaluate switch function */
 353             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 354             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 355             vvdw             = _mm_mul_ps(vvdw,sw);
 356             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 360             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 361             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 362
 363             fscal            = fvdw;
 364
 365             fscal            = _mm_and_ps(fscal,cutoff_mask);
 366
 367             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm_mul_ps(fscal,dx00);
 371             ty               = _mm_mul_ps(fscal,dy00);
 372             tz               = _mm_mul_ps(fscal,dz00);
 373
 374             /* Update vectorial force */
 375             fix0             = _mm_add_ps(fix0,tx);
 376             fiy0             = _mm_add_ps(fiy0,ty);
 377             fiz0             = _mm_add_ps(fiz0,tz);
 378
 379             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 380             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 381             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 382             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 383             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 384
 385             }
 386
 387             /* Inner loop uses 60 flops */
 388         }
 389
 390         /* End of innermost loop */
 391
 392         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 393                                               f+i_coord_offset,fshift+i_shift_offset);
 394
 395         ggid                        = gid[iidx];
 396         /* Update potential energies */
 397         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 398
 399         /* Increment number of inner iterations */
 400         inneriter                  += j_index_end - j_index_start;
 401
 402         /* Outer loop uses 7 flops */
 403     }
 404
 405     /* Increment number of outer iterations */
 406     outeriter        += nri;
 407
 408     /* Update outer/inner flops */
 409
 410     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
 411 }
 412 /*
 413  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 414  * Electrostatics interaction: None
 415  * VdW interaction:            LennardJones
 416  * Geometry:                   Particle-Particle
 417  * Calculate force/pot:        Force
 418  */
 419 void
 420 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 421                     (t_nblist                    * gmx_restrict       nlist,
 422                      rvec                        * gmx_restrict          xx,
 423                      rvec                        * gmx_restrict          ff,
 424                      struct t_forcerec           * gmx_restrict          fr,
 425                      t_mdatoms                   * gmx_restrict     mdatoms,
 426                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 427                      t_nrnb                      * gmx_restrict        nrnb)
 428 {
 429     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 430      * just 0 for non-waters.
 431      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 432      * jnr indices corresponding to data put in the four positions in the SIMD register.
 433      */
 434     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 435     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 436     int              jnrA,jnrB,jnrC,jnrD;
 437     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 438     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 439     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 440     real             rcutoff_scalar;
 441     real             *shiftvec,*fshift,*x,*f;
 442     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 443     real             scratch[4*DIM];
 444     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 445     int              vdwioffset0;
 446     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 447     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 448     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 449     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 450     int              nvdwtype;
 451     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 452     int              *vdwtype;
 453     real             *vdwparam;
 454     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 455     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 456     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 457     real             rswitch_scalar,d_scalar;
 458     __m128           dummy_mask,cutoff_mask;
 459     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 460     __m128           one     = _mm_set1_ps(1.0);
 461     __m128           two     = _mm_set1_ps(2.0);
 462     x                = xx[0];
 463     f                = ff[0];
 464
 465     nri              = nlist->nri;
 466     iinr             = nlist->iinr;
 467     jindex           = nlist->jindex;
 468     jjnr             = nlist->jjnr;
 469     shiftidx         = nlist->shift;
 470     gid              = nlist->gid;
 471     shiftvec         = fr->shift_vec[0];
 472     fshift           = fr->fshift[0];
 473     nvdwtype         = fr->ntype;
 474     vdwparam         = fr->nbfp;
 475     vdwtype          = mdatoms->typeA;
 476
 477     rcutoff_scalar   = fr->ic->rvdw;
 478     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 479     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 480
 481     rswitch_scalar   = fr->ic->rvdw_switch;
 482     rswitch          = _mm_set1_ps(rswitch_scalar);
 483     /* Setup switch parameters */
 484     d_scalar         = rcutoff_scalar-rswitch_scalar;
 485     d                = _mm_set1_ps(d_scalar);
 486     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 487     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 488     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 489     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 490     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 491     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 492
 493     /* Avoid stupid compiler warnings */
 494     jnrA = jnrB = jnrC = jnrD = 0;
 495     j_coord_offsetA = 0;
 496     j_coord_offsetB = 0;
 497     j_coord_offsetC = 0;
 498     j_coord_offsetD = 0;
 499
 500     outeriter        = 0;
 501     inneriter        = 0;
 502
 503     for(iidx=0;iidx<4*DIM;iidx++)
 504     {
 505         scratch[iidx] = 0.0;
 506     }
 507
 508     /* Start outer loop over neighborlists */
 509     for(iidx=0; iidx<nri; iidx++)
 510     {
 511         /* Load shift vector for this list */
 512         i_shift_offset   = DIM*shiftidx[iidx];
 513
 514         /* Load limits for loop over neighbors */
 515         j_index_start    = jindex[iidx];
 516         j_index_end      = jindex[iidx+1];
 517
 518         /* Get outer coordinate index */
 519         inr              = iinr[iidx];
 520         i_coord_offset   = DIM*inr;
 521
 522         /* Load i particle coords and add shift vector */
 523         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 524
 525         fix0             = _mm_setzero_ps();
 526         fiy0             = _mm_setzero_ps();
 527         fiz0             = _mm_setzero_ps();
 528
 529         /* Load parameters for i particles */
 530         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 531
 532         /* Start inner kernel loop */
 533         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 534         {
 535
 536             /* Get j neighbor index, and coordinate index */
 537             jnrA             = jjnr[jidx];
 538             jnrB             = jjnr[jidx+1];
 539             jnrC             = jjnr[jidx+2];
 540             jnrD             = jjnr[jidx+3];
 541             j_coord_offsetA  = DIM*jnrA;
 542             j_coord_offsetB  = DIM*jnrB;
 543             j_coord_offsetC  = DIM*jnrC;
 544             j_coord_offsetD  = DIM*jnrD;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                               x+j_coord_offsetC,x+j_coord_offsetD,
 549                                               &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm_sub_ps(ix0,jx0);
 553             dy00             = _mm_sub_ps(iy0,jy0);
 554             dz00             = _mm_sub_ps(iz0,jz0);
 555
 556             /* Calculate squared distance and things based on it */
 557             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 558
 559             rinv00           = sse2_invsqrt_f(rsq00);
 560
 561             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 562
 563             /* Load parameters for j particles */
 564             vdwjidx0A        = 2*vdwtype[jnrA+0];
 565             vdwjidx0B        = 2*vdwtype[jnrB+0];
 566             vdwjidx0C        = 2*vdwtype[jnrC+0];
 567             vdwjidx0D        = 2*vdwtype[jnrD+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             if (gmx_mm_any_lt(rsq00,rcutoff2))
 574             {
 575
 576             r00              = _mm_mul_ps(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 580                                          vdwparam+vdwioffset0+vdwjidx0B,
 581                                          vdwparam+vdwioffset0+vdwjidx0C,
 582                                          vdwparam+vdwioffset0+vdwjidx0D,
 583                                          &c6_00,&c12_00);
 584
 585             /* LENNARD-JONES DISPERSION/REPULSION */
 586
 587             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 588             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 589             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 590             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 591             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 592
 593             d                = _mm_sub_ps(r00,rswitch);
 594             d                = _mm_max_ps(d,_mm_setzero_ps());
 595             d2               = _mm_mul_ps(d,d);
 596             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)))))));
 597
 598             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 599
 600             /* Evaluate switch function */
 601             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 602             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 603             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 604
 605             fscal            = fvdw;
 606
 607             fscal            = _mm_and_ps(fscal,cutoff_mask);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_ps(fscal,dx00);
 611             ty               = _mm_mul_ps(fscal,dy00);
 612             tz               = _mm_mul_ps(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm_add_ps(fix0,tx);
 616             fiy0             = _mm_add_ps(fiy0,ty);
 617             fiz0             = _mm_add_ps(fiz0,tz);
 618
 619             fjptrA             = f+j_coord_offsetA;
 620             fjptrB             = f+j_coord_offsetB;
 621             fjptrC             = f+j_coord_offsetC;
 622             fjptrD             = f+j_coord_offsetD;
 623             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 624
 625             }
 626
 627             /* Inner loop uses 56 flops */
 628         }
 629
 630         if(jidx<j_index_end)
 631         {
 632
 633             /* Get j neighbor index, and coordinate index */
 634             jnrlistA         = jjnr[jidx];
 635             jnrlistB         = jjnr[jidx+1];
 636             jnrlistC         = jjnr[jidx+2];
 637             jnrlistD         = jjnr[jidx+3];
 638             /* Sign of each element will be negative for non-real atoms.
 639              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 640              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 641              */
 642             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 643             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 644             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 645             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 646             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 647             j_coord_offsetA  = DIM*jnrA;
 648             j_coord_offsetB  = DIM*jnrB;
 649             j_coord_offsetC  = DIM*jnrC;
 650             j_coord_offsetD  = DIM*jnrD;
 651
 652             /* load j atom coordinates */
 653             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 654                                               x+j_coord_offsetC,x+j_coord_offsetD,
 655                                               &jx0,&jy0,&jz0);
 656
 657             /* Calculate displacement vector */
 658             dx00             = _mm_sub_ps(ix0,jx0);
 659             dy00             = _mm_sub_ps(iy0,jy0);
 660             dz00             = _mm_sub_ps(iz0,jz0);
 661
 662             /* Calculate squared distance and things based on it */
 663             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 664
 665             rinv00           = sse2_invsqrt_f(rsq00);
 666
 667             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 668
 669             /* Load parameters for j particles */
 670             vdwjidx0A        = 2*vdwtype[jnrA+0];
 671             vdwjidx0B        = 2*vdwtype[jnrB+0];
 672             vdwjidx0C        = 2*vdwtype[jnrC+0];
 673             vdwjidx0D        = 2*vdwtype[jnrD+0];
 674
 675             /**************************
 676              * CALCULATE INTERACTIONS *
 677              **************************/
 678
 679             if (gmx_mm_any_lt(rsq00,rcutoff2))
 680             {
 681
 682             r00              = _mm_mul_ps(rsq00,rinv00);
 683             r00              = _mm_andnot_ps(dummy_mask,r00);
 684
 685             /* Compute parameters for interactions between i and j atoms */
 686             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 687                                          vdwparam+vdwioffset0+vdwjidx0B,
 688                                          vdwparam+vdwioffset0+vdwjidx0C,
 689                                          vdwparam+vdwioffset0+vdwjidx0D,
 690                                          &c6_00,&c12_00);
 691
 692             /* LENNARD-JONES DISPERSION/REPULSION */
 693
 694             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 695             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 696             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 697             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 698             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 699
 700             d                = _mm_sub_ps(r00,rswitch);
 701             d                = _mm_max_ps(d,_mm_setzero_ps());
 702             d2               = _mm_mul_ps(d,d);
 703             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)))))));
 704
 705             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 706
 707             /* Evaluate switch function */
 708             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 709             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 710             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 711
 712             fscal            = fvdw;
 713
 714             fscal            = _mm_and_ps(fscal,cutoff_mask);
 715
 716             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm_mul_ps(fscal,dx00);
 720             ty               = _mm_mul_ps(fscal,dy00);
 721             tz               = _mm_mul_ps(fscal,dz00);
 722
 723             /* Update vectorial force */
 724             fix0             = _mm_add_ps(fix0,tx);
 725             fiy0             = _mm_add_ps(fiy0,ty);
 726             fiz0             = _mm_add_ps(fiz0,tz);
 727
 728             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 729             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 730             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 731             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 732             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 733
 734             }
 735
 736             /* Inner loop uses 57 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 742                                               f+i_coord_offset,fshift+i_shift_offset);
 743
 744         /* Increment number of inner iterations */
 745         inneriter                  += j_index_end - j_index_start;
 746
 747         /* Outer loop uses 6 flops */
 748     }
 749
 750     /* Increment number of outer iterations */
 751     outeriter        += nri;
 752
 753     /* Update outer/inner flops */
 754
 755     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);
 756 }