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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     int              nvdwtype;
  86     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  90     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  91     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m128d          dummy_mask,cutoff_mask;
  94     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  95     __m128d          one     = _mm_set1_pd(1.0);
  96     __m128d          two     = _mm_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     rcutoff_scalar   = fr->rvdw;
 113     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 114     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 115
 116     rswitch_scalar   = fr->rvdw_switch;
 117     rswitch          = _mm_set1_pd(rswitch_scalar);
 118     /* Setup switch parameters */
 119     d_scalar         = rcutoff_scalar-rswitch_scalar;
 120     d                = _mm_set1_pd(d_scalar);
 121     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 122     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 123     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 124     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 125     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 126     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156
 157         /* Load parameters for i particles */
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         vvdwsum          = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             vdwjidx0A        = 2*vdwtype[jnrA+0];
 191             vdwjidx0B        = 2*vdwtype[jnrB+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             if (gmx_mm_any_lt(rsq00,rcutoff2))
 198             {
 199
 200             r00              = _mm_mul_pd(rsq00,rinv00);
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 204                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 205
 206             /* LENNARD-JONES DISPERSION/REPULSION */
 207
 208             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 209             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 210             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 211             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 212             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 213
 214             d                = _mm_sub_pd(r00,rswitch);
 215             d                = _mm_max_pd(d,_mm_setzero_pd());
 216             d2               = _mm_mul_pd(d,d);
 217             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)))))));
 218
 219             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 220
 221             /* Evaluate switch function */
 222             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 223             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 224             vvdw             = _mm_mul_pd(vvdw,sw);
 225             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 229             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 230
 231             fscal            = fvdw;
 232
 233             fscal            = _mm_and_pd(fscal,cutoff_mask);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm_mul_pd(fscal,dx00);
 237             ty               = _mm_mul_pd(fscal,dy00);
 238             tz               = _mm_mul_pd(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_add_pd(fix0,tx);
 242             fiy0             = _mm_add_pd(fiy0,ty);
 243             fiz0             = _mm_add_pd(fiz0,tz);
 244
 245             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 246
 247             }
 248
 249             /* Inner loop uses 59 flops */
 250         }
 251
 252         if(jidx<j_index_end)
 253         {
 254
 255             jnrA             = jjnr[jidx];
 256             j_coord_offsetA  = DIM*jnrA;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 260                                               &jx0,&jy0,&jz0);
 261
 262             /* Calculate displacement vector */
 263             dx00             = _mm_sub_pd(ix0,jx0);
 264             dy00             = _mm_sub_pd(iy0,jy0);
 265             dz00             = _mm_sub_pd(iz0,jz0);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 269
 270             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 271
 272             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 273
 274             /* Load parameters for j particles */
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             if (gmx_mm_any_lt(rsq00,rcutoff2))
 282             {
 283
 284             r00              = _mm_mul_pd(rsq00,rinv00);
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 288
 289             /* LENNARD-JONES DISPERSION/REPULSION */
 290
 291             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 292             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 293             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 294             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 295             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 296
 297             d                = _mm_sub_pd(r00,rswitch);
 298             d                = _mm_max_pd(d,_mm_setzero_pd());
 299             d2               = _mm_mul_pd(d,d);
 300             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)))))));
 301
 302             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 303
 304             /* Evaluate switch function */
 305             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 306             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 307             vvdw             = _mm_mul_pd(vvdw,sw);
 308             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 312             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 313             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 314
 315             fscal            = fvdw;
 316
 317             fscal            = _mm_and_pd(fscal,cutoff_mask);
 318
 319             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm_mul_pd(fscal,dx00);
 323             ty               = _mm_mul_pd(fscal,dy00);
 324             tz               = _mm_mul_pd(fscal,dz00);
 325
 326             /* Update vectorial force */
 327             fix0             = _mm_add_pd(fix0,tx);
 328             fiy0             = _mm_add_pd(fiy0,ty);
 329             fiz0             = _mm_add_pd(fiz0,tz);
 330
 331             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 332
 333             }
 334
 335             /* Inner loop uses 59 flops */
 336         }
 337
 338         /* End of innermost loop */
 339
 340         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 341                                               f+i_coord_offset,fshift+i_shift_offset);
 342
 343         ggid                        = gid[iidx];
 344         /* Update potential energies */
 345         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 346
 347         /* Increment number of inner iterations */
 348         inneriter                  += j_index_end - j_index_start;
 349
 350         /* Outer loop uses 7 flops */
 351     }
 352
 353     /* Increment number of outer iterations */
 354     outeriter        += nri;
 355
 356     /* Update outer/inner flops */
 357
 358     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 359 }
 360 /*
 361  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 362  * Electrostatics interaction: None
 363  * VdW interaction:            LennardJones
 364  * Geometry:                   Particle-Particle
 365  * Calculate force/pot:        Force
 366  */
 367 void
 368 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_double
 369                     (t_nblist                    * gmx_restrict       nlist,
 370                      rvec                        * gmx_restrict          xx,
 371                      rvec                        * gmx_restrict          ff,
 372                      t_forcerec                  * gmx_restrict          fr,
 373                      t_mdatoms                   * gmx_restrict     mdatoms,
 374                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 375                      t_nrnb                      * gmx_restrict        nrnb)
 376 {
 377     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 378      * just 0 for non-waters.
 379      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 380      * jnr indices corresponding to data put in the four positions in the SIMD register.
 381      */
 382     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 383     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 384     int              jnrA,jnrB;
 385     int              j_coord_offsetA,j_coord_offsetB;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 390     int              vdwioffset0;
 391     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B;
 393     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     int              nvdwtype;
 396     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 397     int              *vdwtype;
 398     real             *vdwparam;
 399     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 400     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 401     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 402     real             rswitch_scalar,d_scalar;
 403     __m128d          dummy_mask,cutoff_mask;
 404     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 405     __m128d          one     = _mm_set1_pd(1.0);
 406     __m128d          two     = _mm_set1_pd(2.0);
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     nvdwtype         = fr->ntype;
 419     vdwparam         = fr->nbfp;
 420     vdwtype          = mdatoms->typeA;
 421
 422     rcutoff_scalar   = fr->rvdw;
 423     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 424     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 425
 426     rswitch_scalar   = fr->rvdw_switch;
 427     rswitch          = _mm_set1_pd(rswitch_scalar);
 428     /* Setup switch parameters */
 429     d_scalar         = rcutoff_scalar-rswitch_scalar;
 430     d                = _mm_set1_pd(d_scalar);
 431     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 432     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 433     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 434     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 435     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 436     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 437
 438     /* Avoid stupid compiler warnings */
 439     jnrA = jnrB = 0;
 440     j_coord_offsetA = 0;
 441     j_coord_offsetB = 0;
 442
 443     outeriter        = 0;
 444     inneriter        = 0;
 445
 446     /* Start outer loop over neighborlists */
 447     for(iidx=0; iidx<nri; iidx++)
 448     {
 449         /* Load shift vector for this list */
 450         i_shift_offset   = DIM*shiftidx[iidx];
 451
 452         /* Load limits for loop over neighbors */
 453         j_index_start    = jindex[iidx];
 454         j_index_end      = jindex[iidx+1];
 455
 456         /* Get outer coordinate index */
 457         inr              = iinr[iidx];
 458         i_coord_offset   = DIM*inr;
 459
 460         /* Load i particle coords and add shift vector */
 461         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 462
 463         fix0             = _mm_setzero_pd();
 464         fiy0             = _mm_setzero_pd();
 465         fiz0             = _mm_setzero_pd();
 466
 467         /* Load parameters for i particles */
 468         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 469
 470         /* Start inner kernel loop */
 471         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 472         {
 473
 474             /* Get j neighbor index, and coordinate index */
 475             jnrA             = jjnr[jidx];
 476             jnrB             = jjnr[jidx+1];
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479
 480             /* load j atom coordinates */
 481             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 482                                               &jx0,&jy0,&jz0);
 483
 484             /* Calculate displacement vector */
 485             dx00             = _mm_sub_pd(ix0,jx0);
 486             dy00             = _mm_sub_pd(iy0,jy0);
 487             dz00             = _mm_sub_pd(iz0,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 491
 492             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 493
 494             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 495
 496             /* Load parameters for j particles */
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm_any_lt(rsq00,rcutoff2))
 505             {
 506
 507             r00              = _mm_mul_pd(rsq00,rinv00);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 512
 513             /* LENNARD-JONES DISPERSION/REPULSION */
 514
 515             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 516             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 517             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 518             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 519             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 520
 521             d                = _mm_sub_pd(r00,rswitch);
 522             d                = _mm_max_pd(d,_mm_setzero_pd());
 523             d2               = _mm_mul_pd(d,d);
 524             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)))))));
 525
 526             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 527
 528             /* Evaluate switch function */
 529             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 530             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 531             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm_and_pd(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm_mul_pd(fscal,dx00);
 539             ty               = _mm_mul_pd(fscal,dy00);
 540             tz               = _mm_mul_pd(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_add_pd(fix0,tx);
 544             fiy0             = _mm_add_pd(fiy0,ty);
 545             fiz0             = _mm_add_pd(fiz0,tz);
 546
 547             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 548
 549             }
 550
 551             /* Inner loop uses 56 flops */
 552         }
 553
 554         if(jidx<j_index_end)
 555         {
 556
 557             jnrA             = jjnr[jidx];
 558             j_coord_offsetA  = DIM*jnrA;
 559
 560             /* load j atom coordinates */
 561             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 562                                               &jx0,&jy0,&jz0);
 563
 564             /* Calculate displacement vector */
 565             dx00             = _mm_sub_pd(ix0,jx0);
 566             dy00             = _mm_sub_pd(iy0,jy0);
 567             dz00             = _mm_sub_pd(iz0,jz0);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 571
 572             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 573
 574             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 575
 576             /* Load parameters for j particles */
 577             vdwjidx0A        = 2*vdwtype[jnrA+0];
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             if (gmx_mm_any_lt(rsq00,rcutoff2))
 584             {
 585
 586             r00              = _mm_mul_pd(rsq00,rinv00);
 587
 588             /* Compute parameters for interactions between i and j atoms */
 589             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 590
 591             /* LENNARD-JONES DISPERSION/REPULSION */
 592
 593             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 594             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 595             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 596             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 597             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 598
 599             d                = _mm_sub_pd(r00,rswitch);
 600             d                = _mm_max_pd(d,_mm_setzero_pd());
 601             d2               = _mm_mul_pd(d,d);
 602             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)))))));
 603
 604             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 605
 606             /* Evaluate switch function */
 607             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 608             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 609             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 610
 611             fscal            = fvdw;
 612
 613             fscal            = _mm_and_pd(fscal,cutoff_mask);
 614
 615             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_pd(fscal,dx00);
 619             ty               = _mm_mul_pd(fscal,dy00);
 620             tz               = _mm_mul_pd(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_add_pd(fix0,tx);
 624             fiy0             = _mm_add_pd(fiy0,ty);
 625             fiz0             = _mm_add_pd(fiz0,tz);
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 628
 629             }
 630
 631             /* Inner loop uses 56 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         /* Increment number of inner iterations */
 640         inneriter                  += j_index_end - j_index_start;
 641
 642         /* Outer loop uses 6 flops */
 643     }
 644
 645     /* Increment number of outer iterations */
 646     outeriter        += nri;
 647
 648     /* Update outer/inner flops */
 649
 650     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*56);
 651 }