src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_double
  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_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  94     real             rswitch_scalar,d_scalar;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(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->rvdw;
 115     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 117
 118     rswitch_scalar   = fr->rvdw_switch;
 119     rswitch          = _mm_set1_pd(rswitch_scalar);
 120     /* Setup switch parameters */
 121     d_scalar         = rcutoff_scalar-rswitch_scalar;
 122     d                = _mm_set1_pd(d_scalar);
 123     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 124     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 127     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 128     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         vvdwsum          = _mm_setzero_pd();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm_sub_pd(ix0,jx0);
 181             dy00             = _mm_sub_pd(iy0,jy0);
 182             dz00             = _mm_sub_pd(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 188
 189             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             r00              = _mm_mul_pd(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 206                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 207
 208             /* LENNARD-JONES DISPERSION/REPULSION */
 209
 210             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 211             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 212             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 213             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 214             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 215
 216             d                = _mm_sub_pd(r00,rswitch);
 217             d                = _mm_max_pd(d,_mm_setzero_pd());
 218             d2               = _mm_mul_pd(d,d);
 219             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 220
 221             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 222
 223             /* Evaluate switch function */
 224             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 225             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 226             vvdw             = _mm_mul_pd(vvdw,sw);
 227             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 231             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 232
 233             fscal            = fvdw;
 234
 235             fscal            = _mm_and_pd(fscal,cutoff_mask);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm_mul_pd(fscal,dx00);
 239             ty               = _mm_mul_pd(fscal,dy00);
 240             tz               = _mm_mul_pd(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_add_pd(fix0,tx);
 244             fiy0             = _mm_add_pd(fiy0,ty);
 245             fiz0             = _mm_add_pd(fiz0,tz);
 246
 247             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 248
 249             }
 250
 251             /* Inner loop uses 59 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             jnrA             = jjnr[jidx];
 258             j_coord_offsetA  = DIM*jnrA;
 259
 260             /* load j atom coordinates */
 261             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 262                                               &jx0,&jy0,&jz0);
 263
 264             /* Calculate displacement vector */
 265             dx00             = _mm_sub_pd(ix0,jx0);
 266             dy00             = _mm_sub_pd(iy0,jy0);
 267             dz00             = _mm_sub_pd(iz0,jz0);
 268
 269             /* Calculate squared distance and things based on it */
 270             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 271
 272             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 273
 274             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 275
 276             /* Load parameters for j particles */
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             if (gmx_mm_any_lt(rsq00,rcutoff2))
 284             {
 285
 286             r00              = _mm_mul_pd(rsq00,rinv00);
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 290
 291             /* LENNARD-JONES DISPERSION/REPULSION */
 292
 293             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 294             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 295             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 296             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 297             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 298
 299             d                = _mm_sub_pd(r00,rswitch);
 300             d                = _mm_max_pd(d,_mm_setzero_pd());
 301             d2               = _mm_mul_pd(d,d);
 302             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 303
 304             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 305
 306             /* Evaluate switch function */
 307             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 308             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 309             vvdw             = _mm_mul_pd(vvdw,sw);
 310             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 314             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 315             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 316
 317             fscal            = fvdw;
 318
 319             fscal            = _mm_and_pd(fscal,cutoff_mask);
 320
 321             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_pd(fscal,dx00);
 325             ty               = _mm_mul_pd(fscal,dy00);
 326             tz               = _mm_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_add_pd(fix0,tx);
 330             fiy0             = _mm_add_pd(fiy0,ty);
 331             fiz0             = _mm_add_pd(fiz0,tz);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 334
 335             }
 336
 337             /* Inner loop uses 59 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 343                                               f+i_coord_offset,fshift+i_shift_offset);
 344
 345         ggid                        = gid[iidx];
 346         /* Update potential energies */
 347         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 348
 349         /* Increment number of inner iterations */
 350         inneriter                  += j_index_end - j_index_start;
 351
 352         /* Outer loop uses 7 flops */
 353     }
 354
 355     /* Increment number of outer iterations */
 356     outeriter        += nri;
 357
 358     /* Update outer/inner flops */
 359
 360     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 361 }
 362 /*
 363  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 364  * Electrostatics interaction: None
 365  * VdW interaction:            LennardJones
 366  * Geometry:                   Particle-Particle
 367  * Calculate force/pot:        Force
 368  */
 369 void
 370 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 371                     (t_nblist                    * gmx_restrict       nlist,
 372                      rvec                        * gmx_restrict          xx,
 373                      rvec                        * gmx_restrict          ff,
 374                      t_forcerec                  * gmx_restrict          fr,
 375                      t_mdatoms                   * gmx_restrict     mdatoms,
 376                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 377                      t_nrnb                      * gmx_restrict        nrnb)
 378 {
 379     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 380      * just 0 for non-waters.
 381      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 382      * jnr indices corresponding to data put in the four positions in the SIMD register.
 383      */
 384     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 385     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 386     int              jnrA,jnrB;
 387     int              j_coord_offsetA,j_coord_offsetB;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 392     int              vdwioffset0;
 393     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 394     int              vdwjidx0A,vdwjidx0B;
 395     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 396     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 397     int              nvdwtype;
 398     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 402     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 403     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 404     real             rswitch_scalar,d_scalar;
 405     __m128d          dummy_mask,cutoff_mask;
 406     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 407     __m128d          one     = _mm_set1_pd(1.0);
 408     __m128d          two     = _mm_set1_pd(2.0);
 409     x                = xx[0];
 410     f                = ff[0];
 411
 412     nri              = nlist->nri;
 413     iinr             = nlist->iinr;
 414     jindex           = nlist->jindex;
 415     jjnr             = nlist->jjnr;
 416     shiftidx         = nlist->shift;
 417     gid              = nlist->gid;
 418     shiftvec         = fr->shift_vec[0];
 419     fshift           = fr->fshift[0];
 420     nvdwtype         = fr->ntype;
 421     vdwparam         = fr->nbfp;
 422     vdwtype          = mdatoms->typeA;
 423
 424     rcutoff_scalar   = fr->rvdw;
 425     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 426     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 427
 428     rswitch_scalar   = fr->rvdw_switch;
 429     rswitch          = _mm_set1_pd(rswitch_scalar);
 430     /* Setup switch parameters */
 431     d_scalar         = rcutoff_scalar-rswitch_scalar;
 432     d                = _mm_set1_pd(d_scalar);
 433     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 434     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 435     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 436     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 437     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 438     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 439
 440     /* Avoid stupid compiler warnings */
 441     jnrA = jnrB = 0;
 442     j_coord_offsetA = 0;
 443     j_coord_offsetB = 0;
 444
 445     outeriter        = 0;
 446     inneriter        = 0;
 447
 448     /* Start outer loop over neighborlists */
 449     for(iidx=0; iidx<nri; iidx++)
 450     {
 451         /* Load shift vector for this list */
 452         i_shift_offset   = DIM*shiftidx[iidx];
 453
 454         /* Load limits for loop over neighbors */
 455         j_index_start    = jindex[iidx];
 456         j_index_end      = jindex[iidx+1];
 457
 458         /* Get outer coordinate index */
 459         inr              = iinr[iidx];
 460         i_coord_offset   = DIM*inr;
 461
 462         /* Load i particle coords and add shift vector */
 463         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 464
 465         fix0             = _mm_setzero_pd();
 466         fiy0             = _mm_setzero_pd();
 467         fiz0             = _mm_setzero_pd();
 468
 469         /* Load parameters for i particles */
 470         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 471
 472         /* Start inner kernel loop */
 473         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 474         {
 475
 476             /* Get j neighbor index, and coordinate index */
 477             jnrA             = jjnr[jidx];
 478             jnrB             = jjnr[jidx+1];
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_pd(ix0,jx0);
 488             dy00             = _mm_sub_pd(iy0,jy0);
 489             dz00             = _mm_sub_pd(iz0,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 493
 494             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 495
 496             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 497
 498             /* Load parameters for j particles */
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm_any_lt(rsq00,rcutoff2))
 507             {
 508
 509             r00              = _mm_mul_pd(rsq00,rinv00);
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 513                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 514
 515             /* LENNARD-JONES DISPERSION/REPULSION */
 516
 517             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 518             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 519             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 520             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 521             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 522
 523             d                = _mm_sub_pd(r00,rswitch);
 524             d                = _mm_max_pd(d,_mm_setzero_pd());
 525             d2               = _mm_mul_pd(d,d);
 526             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 527
 528             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 529
 530             /* Evaluate switch function */
 531             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 532             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 533             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 534
 535             fscal            = fvdw;
 536
 537             fscal            = _mm_and_pd(fscal,cutoff_mask);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_pd(fscal,dx00);
 541             ty               = _mm_mul_pd(fscal,dy00);
 542             tz               = _mm_mul_pd(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_add_pd(fix0,tx);
 546             fiy0             = _mm_add_pd(fiy0,ty);
 547             fiz0             = _mm_add_pd(fiz0,tz);
 548
 549             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 550
 551             }
 552
 553             /* Inner loop uses 56 flops */
 554         }
 555
 556         if(jidx<j_index_end)
 557         {
 558
 559             jnrA             = jjnr[jidx];
 560             j_coord_offsetA  = DIM*jnrA;
 561
 562             /* load j atom coordinates */
 563             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 564                                               &jx0,&jy0,&jz0);
 565
 566             /* Calculate displacement vector */
 567             dx00             = _mm_sub_pd(ix0,jx0);
 568             dy00             = _mm_sub_pd(iy0,jy0);
 569             dz00             = _mm_sub_pd(iz0,jz0);
 570
 571             /* Calculate squared distance and things based on it */
 572             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 573
 574             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 575
 576             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 577
 578             /* Load parameters for j particles */
 579             vdwjidx0A        = 2*vdwtype[jnrA+0];
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             if (gmx_mm_any_lt(rsq00,rcutoff2))
 586             {
 587
 588             r00              = _mm_mul_pd(rsq00,rinv00);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 592
 593             /* LENNARD-JONES DISPERSION/REPULSION */
 594
 595             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 596             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 597             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 598             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 599             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 600
 601             d                = _mm_sub_pd(r00,rswitch);
 602             d                = _mm_max_pd(d,_mm_setzero_pd());
 603             d2               = _mm_mul_pd(d,d);
 604             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 605
 606             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 607
 608             /* Evaluate switch function */
 609             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 610             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 611             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 612
 613             fscal            = fvdw;
 614
 615             fscal            = _mm_and_pd(fscal,cutoff_mask);
 616
 617             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_pd(fscal,dx00);
 621             ty               = _mm_mul_pd(fscal,dy00);
 622             tz               = _mm_mul_pd(fscal,dz00);
 623
 624             /* Update vectorial force */
 625             fix0             = _mm_add_pd(fix0,tx);
 626             fiy0             = _mm_add_pd(fiy0,ty);
 627             fiz0             = _mm_add_pd(fiz0,tz);
 628
 629             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 630
 631             }
 632
 633             /* Inner loop uses 56 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 6 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*56);
 653 }