src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecRFCut_VdwLJSw_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_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     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  96     real             rswitch_scalar,d_scalar;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_pd(fr->ic->k_rf);
 115     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_pd(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 125
 126     rswitch_scalar   = fr->rvdw_switch;
 127     rswitch          = _mm_set1_pd(rswitch_scalar);
 128     /* Setup switch parameters */
 129     d_scalar         = rcutoff_scalar-rswitch_scalar;
 130     d                = _mm_set1_pd(d_scalar);
 131     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 132     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 135     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166
 167         /* Load parameters for i particles */
 168         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 169         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196
 197             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 198
 199             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 203             vdwjidx0A        = 2*vdwtype[jnrA+0];
 204             vdwjidx0B        = 2*vdwtype[jnrB+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             if (gmx_mm_any_lt(rsq00,rcutoff2))
 211             {
 212
 213             r00              = _mm_mul_pd(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_pd(iq0,jq0);
 217             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 219
 220             /* REACTION-FIELD ELECTROSTATICS */
 221             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 222             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 228             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 229             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 230             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 231
 232             d                = _mm_sub_pd(r00,rswitch);
 233             d                = _mm_max_pd(d,_mm_setzero_pd());
 234             d2               = _mm_mul_pd(d,d);
 235             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 236
 237             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 238
 239             /* Evaluate switch function */
 240             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 241             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 242             vvdw             = _mm_mul_pd(vvdw,sw);
 243             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velec            = _mm_and_pd(velec,cutoff_mask);
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 249             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_pd(felec,fvdw);
 252
 253             fscal            = _mm_and_pd(fscal,cutoff_mask);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 257             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 258             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 259
 260             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 261                                                    _mm_mul_pd(dx00,fscal),
 262                                                    _mm_mul_pd(dy00,fscal),
 263                                                    _mm_mul_pd(dz00,fscal));
 264
 265             }
 266
 267             /* Inner loop uses 73 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 289
 290             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = _mm_load_sd(charge+jnrA+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq00,rcutoff2))
 301             {
 302
 303             r00              = _mm_mul_pd(rsq00,rinv00);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq00             = _mm_mul_pd(iq0,jq0);
 307             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 311             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 317             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 318             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 319             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 320
 321             d                = _mm_sub_pd(r00,rswitch);
 322             d                = _mm_max_pd(d,_mm_setzero_pd());
 323             d2               = _mm_mul_pd(d,d);
 324             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 325
 326             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 327
 328             /* Evaluate switch function */
 329             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 330             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 331             vvdw             = _mm_mul_pd(vvdw,sw);
 332             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _mm_and_pd(velec,cutoff_mask);
 336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 339             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 340             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 341
 342             fscal            = _mm_add_pd(felec,fvdw);
 343
 344             fscal            = _mm_and_pd(fscal,cutoff_mask);
 345
 346             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 347
 348             /* Update vectorial force */
 349             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 350             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 351             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 352
 353             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 354                                                    _mm_mul_pd(dx00,fscal),
 355                                                    _mm_mul_pd(dy00,fscal),
 356                                                    _mm_mul_pd(dz00,fscal));
 357
 358             }
 359
 360             /* Inner loop uses 73 flops */
 361         }
 362
 363         /* End of innermost loop */
 364
 365         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 366                                               f+i_coord_offset,fshift+i_shift_offset);
 367
 368         ggid                        = gid[iidx];
 369         /* Update potential energies */
 370         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 371         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 372
 373         /* Increment number of inner iterations */
 374         inneriter                  += j_index_end - j_index_start;
 375
 376         /* Outer loop uses 9 flops */
 377     }
 378
 379     /* Increment number of outer iterations */
 380     outeriter        += nri;
 381
 382     /* Update outer/inner flops */
 383
 384     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 385 }
 386 /*
 387  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 388  * Electrostatics interaction: ReactionField
 389  * VdW interaction:            LennardJones
 390  * Geometry:                   Particle-Particle
 391  * Calculate force/pot:        Force
 392  */
 393 void
 394 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 395                     (t_nblist                    * gmx_restrict       nlist,
 396                      rvec                        * gmx_restrict          xx,
 397                      rvec                        * gmx_restrict          ff,
 398                      t_forcerec                  * gmx_restrict          fr,
 399                      t_mdatoms                   * gmx_restrict     mdatoms,
 400                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 401                      t_nrnb                      * gmx_restrict        nrnb)
 402 {
 403     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 404      * just 0 for non-waters.
 405      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 406      * jnr indices corresponding to data put in the four positions in the SIMD register.
 407      */
 408     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 409     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 410     int              jnrA,jnrB;
 411     int              j_coord_offsetA,j_coord_offsetB;
 412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 413     real             rcutoff_scalar;
 414     real             *shiftvec,*fshift,*x,*f;
 415     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 416     int              vdwioffset0;
 417     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 418     int              vdwjidx0A,vdwjidx0B;
 419     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 420     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 421     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 422     real             *charge;
 423     int              nvdwtype;
 424     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 425     int              *vdwtype;
 426     real             *vdwparam;
 427     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 428     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 429     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 430     real             rswitch_scalar,d_scalar;
 431     __m128d          dummy_mask,cutoff_mask;
 432     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 433     __m128d          one     = _mm_set1_pd(1.0);
 434     __m128d          two     = _mm_set1_pd(2.0);
 435     x                = xx[0];
 436     f                = ff[0];
 437
 438     nri              = nlist->nri;
 439     iinr             = nlist->iinr;
 440     jindex           = nlist->jindex;
 441     jjnr             = nlist->jjnr;
 442     shiftidx         = nlist->shift;
 443     gid              = nlist->gid;
 444     shiftvec         = fr->shift_vec[0];
 445     fshift           = fr->fshift[0];
 446     facel            = _mm_set1_pd(fr->epsfac);
 447     charge           = mdatoms->chargeA;
 448     krf              = _mm_set1_pd(fr->ic->k_rf);
 449     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 450     crf              = _mm_set1_pd(fr->ic->c_rf);
 451     nvdwtype         = fr->ntype;
 452     vdwparam         = fr->nbfp;
 453     vdwtype          = mdatoms->typeA;
 454
 455     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 456     rcutoff_scalar   = fr->rcoulomb;
 457     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 458     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 459
 460     rswitch_scalar   = fr->rvdw_switch;
 461     rswitch          = _mm_set1_pd(rswitch_scalar);
 462     /* Setup switch parameters */
 463     d_scalar         = rcutoff_scalar-rswitch_scalar;
 464     d                = _mm_set1_pd(d_scalar);
 465     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 466     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 467     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 468     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 469     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 470     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 471
 472     /* Avoid stupid compiler warnings */
 473     jnrA = jnrB = 0;
 474     j_coord_offsetA = 0;
 475     j_coord_offsetB = 0;
 476
 477     outeriter        = 0;
 478     inneriter        = 0;
 479
 480     /* Start outer loop over neighborlists */
 481     for(iidx=0; iidx<nri; iidx++)
 482     {
 483         /* Load shift vector for this list */
 484         i_shift_offset   = DIM*shiftidx[iidx];
 485
 486         /* Load limits for loop over neighbors */
 487         j_index_start    = jindex[iidx];
 488         j_index_end      = jindex[iidx+1];
 489
 490         /* Get outer coordinate index */
 491         inr              = iinr[iidx];
 492         i_coord_offset   = DIM*inr;
 493
 494         /* Load i particle coords and add shift vector */
 495         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 496
 497         fix0             = _mm_setzero_pd();
 498         fiy0             = _mm_setzero_pd();
 499         fiz0             = _mm_setzero_pd();
 500
 501         /* Load parameters for i particles */
 502         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 503         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 504
 505         /* Start inner kernel loop */
 506         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 507         {
 508
 509             /* Get j neighbor index, and coordinate index */
 510             jnrA             = jjnr[jidx];
 511             jnrB             = jjnr[jidx+1];
 512             j_coord_offsetA  = DIM*jnrA;
 513             j_coord_offsetB  = DIM*jnrB;
 514
 515             /* load j atom coordinates */
 516             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 517                                               &jx0,&jy0,&jz0);
 518
 519             /* Calculate displacement vector */
 520             dx00             = _mm_sub_pd(ix0,jx0);
 521             dy00             = _mm_sub_pd(iy0,jy0);
 522             dz00             = _mm_sub_pd(iz0,jz0);
 523
 524             /* Calculate squared distance and things based on it */
 525             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 526
 527             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 528
 529             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 530
 531             /* Load parameters for j particles */
 532             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 533             vdwjidx0A        = 2*vdwtype[jnrA+0];
 534             vdwjidx0B        = 2*vdwtype[jnrB+0];
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             if (gmx_mm_any_lt(rsq00,rcutoff2))
 541             {
 542
 543             r00              = _mm_mul_pd(rsq00,rinv00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq00             = _mm_mul_pd(iq0,jq0);
 547             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 548                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 549
 550             /* REACTION-FIELD ELECTROSTATICS */
 551             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 552
 553             /* LENNARD-JONES DISPERSION/REPULSION */
 554
 555             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 556             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 557             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 558             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 559             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 560
 561             d                = _mm_sub_pd(r00,rswitch);
 562             d                = _mm_max_pd(d,_mm_setzero_pd());
 563             d2               = _mm_mul_pd(d,d);
 564             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 565
 566             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 567
 568             /* Evaluate switch function */
 569             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 570             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 571             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 572
 573             fscal            = _mm_add_pd(felec,fvdw);
 574
 575             fscal            = _mm_and_pd(fscal,cutoff_mask);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 579             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 580             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 581
 582             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 583                                                    _mm_mul_pd(dx00,fscal),
 584                                                    _mm_mul_pd(dy00,fscal),
 585                                                    _mm_mul_pd(dz00,fscal));
 586
 587             }
 588
 589             /* Inner loop uses 64 flops */
 590         }
 591
 592         if(jidx<j_index_end)
 593         {
 594
 595             jnrA             = jjnr[jidx];
 596             j_coord_offsetA  = DIM*jnrA;
 597
 598             /* load j atom coordinates */
 599             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 600                                               &jx0,&jy0,&jz0);
 601
 602             /* Calculate displacement vector */
 603             dx00             = _mm_sub_pd(ix0,jx0);
 604             dy00             = _mm_sub_pd(iy0,jy0);
 605             dz00             = _mm_sub_pd(iz0,jz0);
 606
 607             /* Calculate squared distance and things based on it */
 608             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 609
 610             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 611
 612             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 613
 614             /* Load parameters for j particles */
 615             jq0              = _mm_load_sd(charge+jnrA+0);
 616             vdwjidx0A        = 2*vdwtype[jnrA+0];
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm_any_lt(rsq00,rcutoff2))
 623             {
 624
 625             r00              = _mm_mul_pd(rsq00,rinv00);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq00             = _mm_mul_pd(iq0,jq0);
 629             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 630
 631             /* REACTION-FIELD ELECTROSTATICS */
 632             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 633
 634             /* LENNARD-JONES DISPERSION/REPULSION */
 635
 636             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 637             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 638             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 639             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 640             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 641
 642             d                = _mm_sub_pd(r00,rswitch);
 643             d                = _mm_max_pd(d,_mm_setzero_pd());
 644             d2               = _mm_mul_pd(d,d);
 645             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 646
 647             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 648
 649             /* Evaluate switch function */
 650             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 651             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 652             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 653
 654             fscal            = _mm_add_pd(felec,fvdw);
 655
 656             fscal            = _mm_and_pd(fscal,cutoff_mask);
 657
 658             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 659
 660             /* Update vectorial force */
 661             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 662             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 663             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 664
 665             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 666                                                    _mm_mul_pd(dx00,fscal),
 667                                                    _mm_mul_pd(dy00,fscal),
 668                                                    _mm_mul_pd(dz00,fscal));
 669
 670             }
 671
 672             /* Inner loop uses 64 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 678                                               f+i_coord_offset,fshift+i_shift_offset);
 679
 680         /* Increment number of inner iterations */
 681         inneriter                  += j_index_end - j_index_start;
 682
 683         /* Outer loop uses 7 flops */
 684     }
 685
 686     /* Increment number of outer iterations */
 687     outeriter        += nri;
 688
 689     /* Update outer/inner flops */
 690
 691     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 692 }