src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "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_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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           = avx128fma_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_msub_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_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 217
 218             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 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_msub_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             /* Update vectorial force */
 235             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 236             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 237             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 238
 239             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 240                                                    _mm_mul_pd(dx00,fscal),
 241                                                    _mm_mul_pd(dy00,fscal),
 242                                                    _mm_mul_pd(dz00,fscal));
 243
 244             }
 245
 246             /* Inner loop uses 62 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = avx128fma_invsqrt_d(rsq00);
 268
 269             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 270
 271             /* Load parameters for j particles */
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             if (gmx_mm_any_lt(rsq00,rcutoff2))
 279             {
 280
 281             r00              = _mm_mul_pd(rsq00,rinv00);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 290             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 291             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 292             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 293
 294             d                = _mm_sub_pd(r00,rswitch);
 295             d                = _mm_max_pd(d,_mm_setzero_pd());
 296             d2               = _mm_mul_pd(d,d);
 297             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 298
 299             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 300
 301             /* Evaluate switch function */
 302             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 303             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 304             vvdw             = _mm_mul_pd(vvdw,sw);
 305             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 309             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 310             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314             fscal            = _mm_and_pd(fscal,cutoff_mask);
 315
 316             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 317
 318             /* Update vectorial force */
 319             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 320             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 321             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 322
 323             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 324                                                    _mm_mul_pd(dx00,fscal),
 325                                                    _mm_mul_pd(dy00,fscal),
 326                                                    _mm_mul_pd(dz00,fscal));
 327
 328             }
 329
 330             /* Inner loop uses 62 flops */
 331         }
 332
 333         /* End of innermost loop */
 334
 335         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 336                                               f+i_coord_offset,fshift+i_shift_offset);
 337
 338         ggid                        = gid[iidx];
 339         /* Update potential energies */
 340         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 341
 342         /* Increment number of inner iterations */
 343         inneriter                  += j_index_end - j_index_start;
 344
 345         /* Outer loop uses 7 flops */
 346     }
 347
 348     /* Increment number of outer iterations */
 349     outeriter        += nri;
 350
 351     /* Update outer/inner flops */
 352
 353     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 354 }
 355 /*
 356  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 357  * Electrostatics interaction: None
 358  * VdW interaction:            LennardJones
 359  * Geometry:                   Particle-Particle
 360  * Calculate force/pot:        Force
 361  */
 362 void
 363 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 364                     (t_nblist                    * gmx_restrict       nlist,
 365                      rvec                        * gmx_restrict          xx,
 366                      rvec                        * gmx_restrict          ff,
 367                      struct t_forcerec           * gmx_restrict          fr,
 368                      t_mdatoms                   * gmx_restrict     mdatoms,
 369                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 370                      t_nrnb                      * gmx_restrict        nrnb)
 371 {
 372     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 373      * just 0 for non-waters.
 374      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 375      * jnr indices corresponding to data put in the four positions in the SIMD register.
 376      */
 377     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 378     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 379     int              jnrA,jnrB;
 380     int              j_coord_offsetA,j_coord_offsetB;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     int              vdwioffset0;
 386     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B;
 388     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     int              nvdwtype;
 391     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 392     int              *vdwtype;
 393     real             *vdwparam;
 394     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 395     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 396     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 397     real             rswitch_scalar,d_scalar;
 398     __m128d          dummy_mask,cutoff_mask;
 399     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 400     __m128d          one     = _mm_set1_pd(1.0);
 401     __m128d          two     = _mm_set1_pd(2.0);
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     nvdwtype         = fr->ntype;
 414     vdwparam         = fr->nbfp;
 415     vdwtype          = mdatoms->typeA;
 416
 417     rcutoff_scalar   = fr->ic->rvdw;
 418     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 419     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 420
 421     rswitch_scalar   = fr->ic->rvdw_switch;
 422     rswitch          = _mm_set1_pd(rswitch_scalar);
 423     /* Setup switch parameters */
 424     d_scalar         = rcutoff_scalar-rswitch_scalar;
 425     d                = _mm_set1_pd(d_scalar);
 426     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 427     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 428     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 429     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 430     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 431     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 432
 433     /* Avoid stupid compiler warnings */
 434     jnrA = jnrB = 0;
 435     j_coord_offsetA = 0;
 436     j_coord_offsetB = 0;
 437
 438     outeriter        = 0;
 439     inneriter        = 0;
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _mm_setzero_pd();
 459         fiy0             = _mm_setzero_pd();
 460         fiz0             = _mm_setzero_pd();
 461
 462         /* Load parameters for i particles */
 463         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               &jx0,&jy0,&jz0);
 478
 479             /* Calculate displacement vector */
 480             dx00             = _mm_sub_pd(ix0,jx0);
 481             dy00             = _mm_sub_pd(iy0,jy0);
 482             dz00             = _mm_sub_pd(iz0,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 486
 487             rinv00           = avx128fma_invsqrt_d(rsq00);
 488
 489             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 490
 491             /* Load parameters for j particles */
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             if (gmx_mm_any_lt(rsq00,rcutoff2))
 500             {
 501
 502             r00              = _mm_mul_pd(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 507
 508             /* LENNARD-JONES DISPERSION/REPULSION */
 509
 510             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 511             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 512             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 513             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 514             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 515
 516             d                = _mm_sub_pd(r00,rswitch);
 517             d                = _mm_max_pd(d,_mm_setzero_pd());
 518             d2               = _mm_mul_pd(d,d);
 519             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 520
 521             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 522
 523             /* Evaluate switch function */
 524             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 525             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 526             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 527
 528             fscal            = fvdw;
 529
 530             fscal            = _mm_and_pd(fscal,cutoff_mask);
 531
 532             /* Update vectorial force */
 533             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 534             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 535             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 536
 537             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 538                                                    _mm_mul_pd(dx00,fscal),
 539                                                    _mm_mul_pd(dy00,fscal),
 540                                                    _mm_mul_pd(dz00,fscal));
 541
 542             }
 543
 544             /* Inner loop uses 59 flops */
 545         }
 546
 547         if(jidx<j_index_end)
 548         {
 549
 550             jnrA             = jjnr[jidx];
 551             j_coord_offsetA  = DIM*jnrA;
 552
 553             /* load j atom coordinates */
 554             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_pd(ix0,jx0);
 559             dy00             = _mm_sub_pd(iy0,jy0);
 560             dz00             = _mm_sub_pd(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 564
 565             rinv00           = avx128fma_invsqrt_d(rsq00);
 566
 567             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_mm_any_lt(rsq00,rcutoff2))
 577             {
 578
 579             r00              = _mm_mul_pd(rsq00,rinv00);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 583
 584             /* LENNARD-JONES DISPERSION/REPULSION */
 585
 586             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 587             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 588             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 589             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 590             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 591
 592             d                = _mm_sub_pd(r00,rswitch);
 593             d                = _mm_max_pd(d,_mm_setzero_pd());
 594             d2               = _mm_mul_pd(d,d);
 595             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 596
 597             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 598
 599             /* Evaluate switch function */
 600             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 601             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 602             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 603
 604             fscal            = fvdw;
 605
 606             fscal            = _mm_and_pd(fscal,cutoff_mask);
 607
 608             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 609
 610             /* Update vectorial force */
 611             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 612             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 613             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 614
 615             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 616                                                    _mm_mul_pd(dx00,fscal),
 617                                                    _mm_mul_pd(dy00,fscal),
 618                                                    _mm_mul_pd(dz00,fscal));
 619
 620             }
 621
 622             /* Inner loop uses 59 flops */
 623         }
 624
 625         /* End of innermost loop */
 626
 627         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 628                                               f+i_coord_offset,fshift+i_shift_offset);
 629
 630         /* Increment number of inner iterations */
 631         inneriter                  += j_index_end - j_index_start;
 632
 633         /* Outer loop uses 6 flops */
 634     }
 635
 636     /* Increment number of outer iterations */
 637     outeriter        += nri;
 638
 639     /* Update outer/inner flops */
 640
 641     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 642 }