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 #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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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_msub_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_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 218
 219             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 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_msub_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             /* Update vectorial force */
 236             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 237             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 238             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 239
 240             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 241                                                    _mm_mul_pd(dx00,fscal),
 242                                                    _mm_mul_pd(dy00,fscal),
 243                                                    _mm_mul_pd(dz00,fscal));
 244
 245             }
 246
 247             /* Inner loop uses 62 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm_any_lt(rsq00,rcutoff2))
 280             {
 281
 282             r00              = _mm_mul_pd(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 286
 287             /* LENNARD-JONES DISPERSION/REPULSION */
 288
 289             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 290             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 291             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 292             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 293             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 294
 295             d                = _mm_sub_pd(r00,rswitch);
 296             d                = _mm_max_pd(d,_mm_setzero_pd());
 297             d2               = _mm_mul_pd(d,d);
 298             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 299
 300             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 301
 302             /* Evaluate switch function */
 303             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 304             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 305             vvdw             = _mm_mul_pd(vvdw,sw);
 306             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 310             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 311             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 312
 313             fscal            = fvdw;
 314
 315             fscal            = _mm_and_pd(fscal,cutoff_mask);
 316
 317             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 318
 319             /* Update vectorial force */
 320             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 321             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 322             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 323
 324             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 325                                                    _mm_mul_pd(dx00,fscal),
 326                                                    _mm_mul_pd(dy00,fscal),
 327                                                    _mm_mul_pd(dz00,fscal));
 328
 329             }
 330
 331             /* Inner loop uses 62 flops */
 332         }
 333
 334         /* End of innermost loop */
 335
 336         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 337                                               f+i_coord_offset,fshift+i_shift_offset);
 338
 339         ggid                        = gid[iidx];
 340         /* Update potential energies */
 341         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 342
 343         /* Increment number of inner iterations */
 344         inneriter                  += j_index_end - j_index_start;
 345
 346         /* Outer loop uses 7 flops */
 347     }
 348
 349     /* Increment number of outer iterations */
 350     outeriter        += nri;
 351
 352     /* Update outer/inner flops */
 353
 354     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 355 }
 356 /*
 357  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 358  * Electrostatics interaction: None
 359  * VdW interaction:            LennardJones
 360  * Geometry:                   Particle-Particle
 361  * Calculate force/pot:        Force
 362  */
 363 void
 364 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 365                     (t_nblist                    * gmx_restrict       nlist,
 366                      rvec                        * gmx_restrict          xx,
 367                      rvec                        * gmx_restrict          ff,
 368                      t_forcerec                  * gmx_restrict          fr,
 369                      t_mdatoms                   * gmx_restrict     mdatoms,
 370                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 371                      t_nrnb                      * gmx_restrict        nrnb)
 372 {
 373     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 374      * just 0 for non-waters.
 375      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 376      * jnr indices corresponding to data put in the four positions in the SIMD register.
 377      */
 378     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 379     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 380     int              jnrA,jnrB;
 381     int              j_coord_offsetA,j_coord_offsetB;
 382     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 383     real             rcutoff_scalar;
 384     real             *shiftvec,*fshift,*x,*f;
 385     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 386     int              vdwioffset0;
 387     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 388     int              vdwjidx0A,vdwjidx0B;
 389     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 390     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 391     int              nvdwtype;
 392     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 393     int              *vdwtype;
 394     real             *vdwparam;
 395     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 396     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 397     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 398     real             rswitch_scalar,d_scalar;
 399     __m128d          dummy_mask,cutoff_mask;
 400     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 401     __m128d          one     = _mm_set1_pd(1.0);
 402     __m128d          two     = _mm_set1_pd(2.0);
 403     x                = xx[0];
 404     f                = ff[0];
 405
 406     nri              = nlist->nri;
 407     iinr             = nlist->iinr;
 408     jindex           = nlist->jindex;
 409     jjnr             = nlist->jjnr;
 410     shiftidx         = nlist->shift;
 411     gid              = nlist->gid;
 412     shiftvec         = fr->shift_vec[0];
 413     fshift           = fr->fshift[0];
 414     nvdwtype         = fr->ntype;
 415     vdwparam         = fr->nbfp;
 416     vdwtype          = mdatoms->typeA;
 417
 418     rcutoff_scalar   = fr->rvdw;
 419     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 420     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 421
 422     rswitch_scalar   = fr->rvdw_switch;
 423     rswitch          = _mm_set1_pd(rswitch_scalar);
 424     /* Setup switch parameters */
 425     d_scalar         = rcutoff_scalar-rswitch_scalar;
 426     d                = _mm_set1_pd(d_scalar);
 427     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 428     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 429     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 430     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 431     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 432     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 433
 434     /* Avoid stupid compiler warnings */
 435     jnrA = jnrB = 0;
 436     j_coord_offsetA = 0;
 437     j_coord_offsetB = 0;
 438
 439     outeriter        = 0;
 440     inneriter        = 0;
 441
 442     /* Start outer loop over neighborlists */
 443     for(iidx=0; iidx<nri; iidx++)
 444     {
 445         /* Load shift vector for this list */
 446         i_shift_offset   = DIM*shiftidx[iidx];
 447
 448         /* Load limits for loop over neighbors */
 449         j_index_start    = jindex[iidx];
 450         j_index_end      = jindex[iidx+1];
 451
 452         /* Get outer coordinate index */
 453         inr              = iinr[iidx];
 454         i_coord_offset   = DIM*inr;
 455
 456         /* Load i particle coords and add shift vector */
 457         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 458
 459         fix0             = _mm_setzero_pd();
 460         fiy0             = _mm_setzero_pd();
 461         fiz0             = _mm_setzero_pd();
 462
 463         /* Load parameters for i particles */
 464         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 465
 466         /* Start inner kernel loop */
 467         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrA             = jjnr[jidx];
 472             jnrB             = jjnr[jidx+1];
 473             j_coord_offsetA  = DIM*jnrA;
 474             j_coord_offsetB  = DIM*jnrB;
 475
 476             /* load j atom coordinates */
 477             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                               &jx0,&jy0,&jz0);
 479
 480             /* Calculate displacement vector */
 481             dx00             = _mm_sub_pd(ix0,jx0);
 482             dy00             = _mm_sub_pd(iy0,jy0);
 483             dz00             = _mm_sub_pd(iz0,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 487
 488             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 489
 490             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 491
 492             /* Load parameters for j particles */
 493             vdwjidx0A        = 2*vdwtype[jnrA+0];
 494             vdwjidx0B        = 2*vdwtype[jnrB+0];
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             if (gmx_mm_any_lt(rsq00,rcutoff2))
 501             {
 502
 503             r00              = _mm_mul_pd(rsq00,rinv00);
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 508
 509             /* LENNARD-JONES DISPERSION/REPULSION */
 510
 511             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 512             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 513             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 514             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 515             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 516
 517             d                = _mm_sub_pd(r00,rswitch);
 518             d                = _mm_max_pd(d,_mm_setzero_pd());
 519             d2               = _mm_mul_pd(d,d);
 520             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 521
 522             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 523
 524             /* Evaluate switch function */
 525             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 526             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 527             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 528
 529             fscal            = fvdw;
 530
 531             fscal            = _mm_and_pd(fscal,cutoff_mask);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 535             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 536             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 537
 538             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 539                                                    _mm_mul_pd(dx00,fscal),
 540                                                    _mm_mul_pd(dy00,fscal),
 541                                                    _mm_mul_pd(dz00,fscal));
 542
 543             }
 544
 545             /* Inner loop uses 59 flops */
 546         }
 547
 548         if(jidx<j_index_end)
 549         {
 550
 551             jnrA             = jjnr[jidx];
 552             j_coord_offsetA  = DIM*jnrA;
 553
 554             /* load j atom coordinates */
 555             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 556                                               &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx00             = _mm_sub_pd(ix0,jx0);
 560             dy00             = _mm_sub_pd(iy0,jy0);
 561             dz00             = _mm_sub_pd(iz0,jz0);
 562
 563             /* Calculate squared distance and things based on it */
 564             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 565
 566             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 567
 568             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 569
 570             /* Load parameters for j particles */
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq00,rcutoff2))
 578             {
 579
 580             r00              = _mm_mul_pd(rsq00,rinv00);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 584
 585             /* LENNARD-JONES DISPERSION/REPULSION */
 586
 587             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 588             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 589             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 590             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 591             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 592
 593             d                = _mm_sub_pd(r00,rswitch);
 594             d                = _mm_max_pd(d,_mm_setzero_pd());
 595             d2               = _mm_mul_pd(d,d);
 596             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 597
 598             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 599
 600             /* Evaluate switch function */
 601             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 602             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 603             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 604
 605             fscal            = fvdw;
 606
 607             fscal            = _mm_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 610
 611             /* Update vectorial force */
 612             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 613             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 614             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 615
 616             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 617                                                    _mm_mul_pd(dx00,fscal),
 618                                                    _mm_mul_pd(dy00,fscal),
 619                                                    _mm_mul_pd(dz00,fscal));
 620
 621             }
 622
 623             /* Inner loop uses 59 flops */
 624         }
 625
 626         /* End of innermost loop */
 627
 628         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 629                                               f+i_coord_offset,fshift+i_shift_offset);
 630
 631         /* Increment number of inner iterations */
 632         inneriter                  += j_index_end - j_index_start;
 633
 634         /* Outer loop uses 6 flops */
 635     }
 636
 637     /* Increment number of outer iterations */
 638     outeriter        += nri;
 639
 640     /* Update outer/inner flops */
 641
 642     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 643 }