src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_avx_128_fma_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 avx_128_fma_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 "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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_msub_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_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 220
 221             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 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_msub_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             /* Update vectorial force */
 238             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 239             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 240             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 243                                                    _mm_mul_pd(dx00,fscal),
 244                                                    _mm_mul_pd(dy00,fscal),
 245                                                    _mm_mul_pd(dz00,fscal));
 246
 247             }
 248
 249             /* Inner loop uses 62 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_msub_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_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 301
 302             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 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_msub_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             /* Update vectorial force */
 322             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 323             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 324             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 325
 326             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 327                                                    _mm_mul_pd(dx00,fscal),
 328                                                    _mm_mul_pd(dy00,fscal),
 329                                                    _mm_mul_pd(dz00,fscal));
 330
 331             }
 332
 333             /* Inner loop uses 62 flops */
 334         }
 335
 336         /* End of innermost loop */
 337
 338         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 339                                               f+i_coord_offset,fshift+i_shift_offset);
 340
 341         ggid                        = gid[iidx];
 342         /* Update potential energies */
 343         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 344
 345         /* Increment number of inner iterations */
 346         inneriter                  += j_index_end - j_index_start;
 347
 348         /* Outer loop uses 7 flops */
 349     }
 350
 351     /* Increment number of outer iterations */
 352     outeriter        += nri;
 353
 354     /* Update outer/inner flops */
 355
 356     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 357 }
 358 /*
 359  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 360  * Electrostatics interaction: None
 361  * VdW interaction:            LennardJones
 362  * Geometry:                   Particle-Particle
 363  * Calculate force/pot:        Force
 364  */
 365 void
 366 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 367                     (t_nblist                    * gmx_restrict       nlist,
 368                      rvec                        * gmx_restrict          xx,
 369                      rvec                        * gmx_restrict          ff,
 370                      t_forcerec                  * gmx_restrict          fr,
 371                      t_mdatoms                   * gmx_restrict     mdatoms,
 372                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 373                      t_nrnb                      * gmx_restrict        nrnb)
 374 {
 375     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 376      * just 0 for non-waters.
 377      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 378      * jnr indices corresponding to data put in the four positions in the SIMD register.
 379      */
 380     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 381     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 382     int              jnrA,jnrB;
 383     int              j_coord_offsetA,j_coord_offsetB;
 384     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 385     real             rcutoff_scalar;
 386     real             *shiftvec,*fshift,*x,*f;
 387     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 388     int              vdwioffset0;
 389     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 390     int              vdwjidx0A,vdwjidx0B;
 391     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 392     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 393     int              nvdwtype;
 394     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 395     int              *vdwtype;
 396     real             *vdwparam;
 397     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 398     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 399     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 400     real             rswitch_scalar,d_scalar;
 401     __m128d          dummy_mask,cutoff_mask;
 402     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 403     __m128d          one     = _mm_set1_pd(1.0);
 404     __m128d          two     = _mm_set1_pd(2.0);
 405     x                = xx[0];
 406     f                = ff[0];
 407
 408     nri              = nlist->nri;
 409     iinr             = nlist->iinr;
 410     jindex           = nlist->jindex;
 411     jjnr             = nlist->jjnr;
 412     shiftidx         = nlist->shift;
 413     gid              = nlist->gid;
 414     shiftvec         = fr->shift_vec[0];
 415     fshift           = fr->fshift[0];
 416     nvdwtype         = fr->ntype;
 417     vdwparam         = fr->nbfp;
 418     vdwtype          = mdatoms->typeA;
 419
 420     rcutoff_scalar   = fr->rvdw;
 421     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 422     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 423
 424     rswitch_scalar   = fr->rvdw_switch;
 425     rswitch          = _mm_set1_pd(rswitch_scalar);
 426     /* Setup switch parameters */
 427     d_scalar         = rcutoff_scalar-rswitch_scalar;
 428     d                = _mm_set1_pd(d_scalar);
 429     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 430     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 431     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 432     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 433     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 434     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 435
 436     /* Avoid stupid compiler warnings */
 437     jnrA = jnrB = 0;
 438     j_coord_offsetA = 0;
 439     j_coord_offsetB = 0;
 440
 441     outeriter        = 0;
 442     inneriter        = 0;
 443
 444     /* Start outer loop over neighborlists */
 445     for(iidx=0; iidx<nri; iidx++)
 446     {
 447         /* Load shift vector for this list */
 448         i_shift_offset   = DIM*shiftidx[iidx];
 449
 450         /* Load limits for loop over neighbors */
 451         j_index_start    = jindex[iidx];
 452         j_index_end      = jindex[iidx+1];
 453
 454         /* Get outer coordinate index */
 455         inr              = iinr[iidx];
 456         i_coord_offset   = DIM*inr;
 457
 458         /* Load i particle coords and add shift vector */
 459         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 460
 461         fix0             = _mm_setzero_pd();
 462         fiy0             = _mm_setzero_pd();
 463         fiz0             = _mm_setzero_pd();
 464
 465         /* Load parameters for i particles */
 466         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 467
 468         /* Start inner kernel loop */
 469         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrA             = jjnr[jidx];
 474             jnrB             = jjnr[jidx+1];
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477
 478             /* load j atom coordinates */
 479             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                               &jx0,&jy0,&jz0);
 481
 482             /* Calculate displacement vector */
 483             dx00             = _mm_sub_pd(ix0,jx0);
 484             dy00             = _mm_sub_pd(iy0,jy0);
 485             dz00             = _mm_sub_pd(iz0,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 489
 490             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 491
 492             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 493
 494             /* Load parameters for j particles */
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496             vdwjidx0B        = 2*vdwtype[jnrB+0];
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm_any_lt(rsq00,rcutoff2))
 503             {
 504
 505             r00              = _mm_mul_pd(rsq00,rinv00);
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 509                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 510
 511             /* LENNARD-JONES DISPERSION/REPULSION */
 512
 513             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 514             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 515             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 516             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 517             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 518
 519             d                = _mm_sub_pd(r00,rswitch);
 520             d                = _mm_max_pd(d,_mm_setzero_pd());
 521             d2               = _mm_mul_pd(d,d);
 522             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 523
 524             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 525
 526             /* Evaluate switch function */
 527             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 528             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 529             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 530
 531             fscal            = fvdw;
 532
 533             fscal            = _mm_and_pd(fscal,cutoff_mask);
 534
 535             /* Update vectorial force */
 536             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 537             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 538             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 539
 540             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 541                                                    _mm_mul_pd(dx00,fscal),
 542                                                    _mm_mul_pd(dy00,fscal),
 543                                                    _mm_mul_pd(dz00,fscal));
 544
 545             }
 546
 547             /* Inner loop uses 59 flops */
 548         }
 549
 550         if(jidx<j_index_end)
 551         {
 552
 553             jnrA             = jjnr[jidx];
 554             j_coord_offsetA  = DIM*jnrA;
 555
 556             /* load j atom coordinates */
 557             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 558                                               &jx0,&jy0,&jz0);
 559
 560             /* Calculate displacement vector */
 561             dx00             = _mm_sub_pd(ix0,jx0);
 562             dy00             = _mm_sub_pd(iy0,jy0);
 563             dz00             = _mm_sub_pd(iz0,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 567
 568             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 569
 570             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 571
 572             /* Load parameters for j particles */
 573             vdwjidx0A        = 2*vdwtype[jnrA+0];
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq00,rcutoff2))
 580             {
 581
 582             r00              = _mm_mul_pd(rsq00,rinv00);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 586
 587             /* LENNARD-JONES DISPERSION/REPULSION */
 588
 589             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 590             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 591             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 592             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 593             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 594
 595             d                = _mm_sub_pd(r00,rswitch);
 596             d                = _mm_max_pd(d,_mm_setzero_pd());
 597             d2               = _mm_mul_pd(d,d);
 598             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 599
 600             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 601
 602             /* Evaluate switch function */
 603             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 604             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 605             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 606
 607             fscal            = fvdw;
 608
 609             fscal            = _mm_and_pd(fscal,cutoff_mask);
 610
 611             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 612
 613             /* Update vectorial force */
 614             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 615             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 616             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 617
 618             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 619                                                    _mm_mul_pd(dx00,fscal),
 620                                                    _mm_mul_pd(dy00,fscal),
 621                                                    _mm_mul_pd(dz00,fscal));
 622
 623             }
 624
 625             /* Inner loop uses 59 flops */
 626         }
 627
 628         /* End of innermost loop */
 629
 630         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 631                                               f+i_coord_offset,fshift+i_shift_offset);
 632
 633         /* Increment number of inner iterations */
 634         inneriter                  += j_index_end - j_index_start;
 635
 636         /* Outer loop uses 6 flops */
 637     }
 638
 639     /* Increment number of outer iterations */
 640     outeriter        += nri;
 641
 642     /* Update outer/inner flops */
 643
 644     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 645 }