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) 2012,2013,2014,2015,2017, 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 #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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_double
  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_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  89     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  90     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     rcutoff_scalar   = fr->ic->rvdw;
 112     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 114
 115     rswitch_scalar   = fr->ic->rvdw_switch;
 116     rswitch          = _mm_set1_pd(rswitch_scalar);
 117     /* Setup switch parameters */
 118     d_scalar         = rcutoff_scalar-rswitch_scalar;
 119     d                = _mm_set1_pd(d_scalar);
 120     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 121     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 122     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 123     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 124     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155
 156         /* Load parameters for i particles */
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         vvdwsum          = _mm_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_pd(ix0,jx0);
 178             dy00             = _mm_sub_pd(iy0,jy0);
 179             dz00             = _mm_sub_pd(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 183
 184             rinv00           = sse41_invsqrt_d(rsq00);
 185
 186             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             if (gmx_mm_any_lt(rsq00,rcutoff2))
 197             {
 198
 199             r00              = _mm_mul_pd(rsq00,rinv00);
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 204
 205             /* LENNARD-JONES DISPERSION/REPULSION */
 206
 207             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 208             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 209             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 210             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 211             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 212
 213             d                = _mm_sub_pd(r00,rswitch);
 214             d                = _mm_max_pd(d,_mm_setzero_pd());
 215             d2               = _mm_mul_pd(d,d);
 216             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)))))));
 217
 218             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 219
 220             /* Evaluate switch function */
 221             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 222             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 223             vvdw             = _mm_mul_pd(vvdw,sw);
 224             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 228             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             fscal            = _mm_and_pd(fscal,cutoff_mask);
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_pd(fscal,dx00);
 236             ty               = _mm_mul_pd(fscal,dy00);
 237             tz               = _mm_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_pd(fix0,tx);
 241             fiy0             = _mm_add_pd(fiy0,ty);
 242             fiz0             = _mm_add_pd(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 245
 246             }
 247
 248             /* Inner loop uses 59 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             jnrA             = jjnr[jidx];
 255             j_coord_offsetA  = DIM*jnrA;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 259                                               &jx0,&jy0,&jz0);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm_sub_pd(ix0,jx0);
 263             dy00             = _mm_sub_pd(iy0,jy0);
 264             dz00             = _mm_sub_pd(iz0,jz0);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 268
 269             rinv00           = sse41_invsqrt_d(rsq00);
 270
 271             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 272
 273             /* Load parameters for j particles */
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             r00              = _mm_mul_pd(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 292             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 293             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 294             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 295
 296             d                = _mm_sub_pd(r00,rswitch);
 297             d                = _mm_max_pd(d,_mm_setzero_pd());
 298             d2               = _mm_mul_pd(d,d);
 299             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)))))));
 300
 301             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 302
 303             /* Evaluate switch function */
 304             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 305             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 306             vvdw             = _mm_mul_pd(vvdw,sw);
 307             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 311             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 312             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 313
 314             fscal            = fvdw;
 315
 316             fscal            = _mm_and_pd(fscal,cutoff_mask);
 317
 318             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_pd(fscal,dx00);
 322             ty               = _mm_mul_pd(fscal,dy00);
 323             tz               = _mm_mul_pd(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm_add_pd(fix0,tx);
 327             fiy0             = _mm_add_pd(fiy0,ty);
 328             fiz0             = _mm_add_pd(fiz0,tz);
 329
 330             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 331
 332             }
 333
 334             /* Inner loop uses 59 flops */
 335         }
 336
 337         /* End of innermost loop */
 338
 339         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 340                                               f+i_coord_offset,fshift+i_shift_offset);
 341
 342         ggid                        = gid[iidx];
 343         /* Update potential energies */
 344         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 345
 346         /* Increment number of inner iterations */
 347         inneriter                  += j_index_end - j_index_start;
 348
 349         /* Outer loop uses 7 flops */
 350     }
 351
 352     /* Increment number of outer iterations */
 353     outeriter        += nri;
 354
 355     /* Update outer/inner flops */
 356
 357     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 358 }
 359 /*
 360  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 361  * Electrostatics interaction: None
 362  * VdW interaction:            LennardJones
 363  * Geometry:                   Particle-Particle
 364  * Calculate force/pot:        Force
 365  */
 366 void
 367 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 368                     (t_nblist                    * gmx_restrict       nlist,
 369                      rvec                        * gmx_restrict          xx,
 370                      rvec                        * gmx_restrict          ff,
 371                      struct t_forcerec           * gmx_restrict          fr,
 372                      t_mdatoms                   * gmx_restrict     mdatoms,
 373                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 374                      t_nrnb                      * gmx_restrict        nrnb)
 375 {
 376     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 377      * just 0 for non-waters.
 378      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 379      * jnr indices corresponding to data put in the four positions in the SIMD register.
 380      */
 381     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 382     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 383     int              jnrA,jnrB;
 384     int              j_coord_offsetA,j_coord_offsetB;
 385     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 386     real             rcutoff_scalar;
 387     real             *shiftvec,*fshift,*x,*f;
 388     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 389     int              vdwioffset0;
 390     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 391     int              vdwjidx0A,vdwjidx0B;
 392     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 393     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 394     int              nvdwtype;
 395     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 396     int              *vdwtype;
 397     real             *vdwparam;
 398     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 399     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 400     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 401     real             rswitch_scalar,d_scalar;
 402     __m128d          dummy_mask,cutoff_mask;
 403     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 404     __m128d          one     = _mm_set1_pd(1.0);
 405     __m128d          two     = _mm_set1_pd(2.0);
 406     x                = xx[0];
 407     f                = ff[0];
 408
 409     nri              = nlist->nri;
 410     iinr             = nlist->iinr;
 411     jindex           = nlist->jindex;
 412     jjnr             = nlist->jjnr;
 413     shiftidx         = nlist->shift;
 414     gid              = nlist->gid;
 415     shiftvec         = fr->shift_vec[0];
 416     fshift           = fr->fshift[0];
 417     nvdwtype         = fr->ntype;
 418     vdwparam         = fr->nbfp;
 419     vdwtype          = mdatoms->typeA;
 420
 421     rcutoff_scalar   = fr->ic->rvdw;
 422     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 423     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 424
 425     rswitch_scalar   = fr->ic->rvdw_switch;
 426     rswitch          = _mm_set1_pd(rswitch_scalar);
 427     /* Setup switch parameters */
 428     d_scalar         = rcutoff_scalar-rswitch_scalar;
 429     d                = _mm_set1_pd(d_scalar);
 430     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 431     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 432     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 433     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 434     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 435     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 436
 437     /* Avoid stupid compiler warnings */
 438     jnrA = jnrB = 0;
 439     j_coord_offsetA = 0;
 440     j_coord_offsetB = 0;
 441
 442     outeriter        = 0;
 443     inneriter        = 0;
 444
 445     /* Start outer loop over neighborlists */
 446     for(iidx=0; iidx<nri; iidx++)
 447     {
 448         /* Load shift vector for this list */
 449         i_shift_offset   = DIM*shiftidx[iidx];
 450
 451         /* Load limits for loop over neighbors */
 452         j_index_start    = jindex[iidx];
 453         j_index_end      = jindex[iidx+1];
 454
 455         /* Get outer coordinate index */
 456         inr              = iinr[iidx];
 457         i_coord_offset   = DIM*inr;
 458
 459         /* Load i particle coords and add shift vector */
 460         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 461
 462         fix0             = _mm_setzero_pd();
 463         fiy0             = _mm_setzero_pd();
 464         fiz0             = _mm_setzero_pd();
 465
 466         /* Load parameters for i particles */
 467         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 468
 469         /* Start inner kernel loop */
 470         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrA             = jjnr[jidx];
 475             jnrB             = jjnr[jidx+1];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_pd(ix0,jx0);
 485             dy00             = _mm_sub_pd(iy0,jy0);
 486             dz00             = _mm_sub_pd(iz0,jz0);
 487
 488             /* Calculate squared distance and things based on it */
 489             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 490
 491             rinv00           = sse41_invsqrt_d(rsq00);
 492
 493             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 494
 495             /* Load parameters for j particles */
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497             vdwjidx0B        = 2*vdwtype[jnrB+0];
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             if (gmx_mm_any_lt(rsq00,rcutoff2))
 504             {
 505
 506             r00              = _mm_mul_pd(rsq00,rinv00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 510                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 515             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 516             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 517             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 518             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 519
 520             d                = _mm_sub_pd(r00,rswitch);
 521             d                = _mm_max_pd(d,_mm_setzero_pd());
 522             d2               = _mm_mul_pd(d,d);
 523             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)))))));
 524
 525             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 526
 527             /* Evaluate switch function */
 528             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 529             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 530             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 531
 532             fscal            = fvdw;
 533
 534             fscal            = _mm_and_pd(fscal,cutoff_mask);
 535
 536             /* Calculate temporary vectorial force */
 537             tx               = _mm_mul_pd(fscal,dx00);
 538             ty               = _mm_mul_pd(fscal,dy00);
 539             tz               = _mm_mul_pd(fscal,dz00);
 540
 541             /* Update vectorial force */
 542             fix0             = _mm_add_pd(fix0,tx);
 543             fiy0             = _mm_add_pd(fiy0,ty);
 544             fiz0             = _mm_add_pd(fiz0,tz);
 545
 546             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 547
 548             }
 549
 550             /* Inner loop uses 56 flops */
 551         }
 552
 553         if(jidx<j_index_end)
 554         {
 555
 556             jnrA             = jjnr[jidx];
 557             j_coord_offsetA  = DIM*jnrA;
 558
 559             /* load j atom coordinates */
 560             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 561                                               &jx0,&jy0,&jz0);
 562
 563             /* Calculate displacement vector */
 564             dx00             = _mm_sub_pd(ix0,jx0);
 565             dy00             = _mm_sub_pd(iy0,jy0);
 566             dz00             = _mm_sub_pd(iz0,jz0);
 567
 568             /* Calculate squared distance and things based on it */
 569             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 570
 571             rinv00           = sse41_invsqrt_d(rsq00);
 572
 573             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 574
 575             /* Load parameters for j particles */
 576             vdwjidx0A        = 2*vdwtype[jnrA+0];
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             if (gmx_mm_any_lt(rsq00,rcutoff2))
 583             {
 584
 585             r00              = _mm_mul_pd(rsq00,rinv00);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 589
 590             /* LENNARD-JONES DISPERSION/REPULSION */
 591
 592             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 593             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 594             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 595             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 596             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 597
 598             d                = _mm_sub_pd(r00,rswitch);
 599             d                = _mm_max_pd(d,_mm_setzero_pd());
 600             d2               = _mm_mul_pd(d,d);
 601             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)))))));
 602
 603             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 604
 605             /* Evaluate switch function */
 606             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 607             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 608             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 609
 610             fscal            = fvdw;
 611
 612             fscal            = _mm_and_pd(fscal,cutoff_mask);
 613
 614             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_pd(fscal,dx00);
 618             ty               = _mm_mul_pd(fscal,dy00);
 619             tz               = _mm_mul_pd(fscal,dz00);
 620
 621             /* Update vectorial force */
 622             fix0             = _mm_add_pd(fix0,tx);
 623             fiy0             = _mm_add_pd(fiy0,ty);
 624             fiz0             = _mm_add_pd(fiz0,tz);
 625
 626             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 627
 628             }
 629
 630             /* Inner loop uses 56 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         /* Increment number of inner iterations */
 639         inneriter                  += j_index_end - j_index_start;
 640
 641         /* Outer loop uses 6 flops */
 642     }
 643
 644     /* Increment number of outer iterations */
 645     outeriter        += nri;
 646
 647     /* Update outer/inner flops */
 648
 649     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*56);
 650 }