src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sse2_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 sse2_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_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_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     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rcoulomb_switch;
 132     rswitch          = _mm_set1_pd(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm_set1_pd(d_scalar);
 136     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _mm_setzero_pd();
 169         fiy0             = _mm_setzero_pd();
 170         fiz0             = _mm_setzero_pd();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 174         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 203
 204             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_mm_any_lt(rsq00,rcutoff2))
 216             {
 217
 218             r00              = _mm_mul_pd(rsq00,rinv00);
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_pd(iq0,jq0);
 222             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 224
 225             /* EWALD ELECTROSTATICS */
 226
 227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 228             ewrt             = _mm_mul_pd(r00,ewtabscale);
 229             ewitab           = _mm_cvttpd_epi32(ewrt);
 230             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 231             ewitab           = _mm_slli_epi32(ewitab,2);
 232             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 233             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 234             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 235             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 236             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 237             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 238             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 239             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 240             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 241             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 248             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 250
 251             d                = _mm_sub_pd(r00,rswitch);
 252             d                = _mm_max_pd(d,_mm_setzero_pd());
 253             d2               = _mm_mul_pd(d,d);
 254             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 255
 256             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 257
 258             /* Evaluate switch function */
 259             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 260             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 261             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 262             velec            = _mm_mul_pd(velec,sw);
 263             vvdw             = _mm_mul_pd(vvdw,sw);
 264             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velec            = _mm_and_pd(velec,cutoff_mask);
 268             velecsum         = _mm_add_pd(velecsum,velec);
 269             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 270             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_pd(felec,fvdw);
 273
 274             fscal            = _mm_and_pd(fscal,cutoff_mask);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_pd(fscal,dx00);
 278             ty               = _mm_mul_pd(fscal,dy00);
 279             tz               = _mm_mul_pd(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_add_pd(fix0,tx);
 283             fiy0             = _mm_add_pd(fiy0,ty);
 284             fiz0             = _mm_add_pd(fiz0,tz);
 285
 286             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 287
 288             }
 289
 290             /* Inner loop uses 83 flops */
 291         }
 292
 293         if(jidx<j_index_end)
 294         {
 295
 296             jnrA             = jjnr[jidx];
 297             j_coord_offsetA  = DIM*jnrA;
 298
 299             /* load j atom coordinates */
 300             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 301                                               &jx0,&jy0,&jz0);
 302
 303             /* Calculate displacement vector */
 304             dx00             = _mm_sub_pd(ix0,jx0);
 305             dy00             = _mm_sub_pd(iy0,jy0);
 306             dz00             = _mm_sub_pd(iz0,jz0);
 307
 308             /* Calculate squared distance and things based on it */
 309             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 310
 311             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 312
 313             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 314
 315             /* Load parameters for j particles */
 316             jq0              = _mm_load_sd(charge+jnrA+0);
 317             vdwjidx0A        = 2*vdwtype[jnrA+0];
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq00,rcutoff2))
 324             {
 325
 326             r00              = _mm_mul_pd(rsq00,rinv00);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _mm_mul_pd(iq0,jq0);
 330             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 331
 332             /* EWALD ELECTROSTATICS */
 333
 334             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 335             ewrt             = _mm_mul_pd(r00,ewtabscale);
 336             ewitab           = _mm_cvttpd_epi32(ewrt);
 337             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 338             ewitab           = _mm_slli_epi32(ewitab,2);
 339             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 340             ewtabD           = _mm_setzero_pd();
 341             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 342             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 343             ewtabFn          = _mm_setzero_pd();
 344             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 345             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 346             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 347             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 348             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 349
 350             /* LENNARD-JONES DISPERSION/REPULSION */
 351
 352             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 353             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 354             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 355             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 356             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 357
 358             d                = _mm_sub_pd(r00,rswitch);
 359             d                = _mm_max_pd(d,_mm_setzero_pd());
 360             d2               = _mm_mul_pd(d,d);
 361             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 362
 363             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 364
 365             /* Evaluate switch function */
 366             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 367             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 368             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 369             velec            = _mm_mul_pd(velec,sw);
 370             vvdw             = _mm_mul_pd(vvdw,sw);
 371             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_pd(velec,cutoff_mask);
 375             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 376             velecsum         = _mm_add_pd(velecsum,velec);
 377             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 378             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 379             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 380
 381             fscal            = _mm_add_pd(felec,fvdw);
 382
 383             fscal            = _mm_and_pd(fscal,cutoff_mask);
 384
 385             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm_mul_pd(fscal,dx00);
 389             ty               = _mm_mul_pd(fscal,dy00);
 390             tz               = _mm_mul_pd(fscal,dz00);
 391
 392             /* Update vectorial force */
 393             fix0             = _mm_add_pd(fix0,tx);
 394             fiy0             = _mm_add_pd(fiy0,ty);
 395             fiz0             = _mm_add_pd(fiz0,tz);
 396
 397             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 398
 399             }
 400
 401             /* Inner loop uses 83 flops */
 402         }
 403
 404         /* End of innermost loop */
 405
 406         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 407                                               f+i_coord_offset,fshift+i_shift_offset);
 408
 409         ggid                        = gid[iidx];
 410         /* Update potential energies */
 411         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 412         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 413
 414         /* Increment number of inner iterations */
 415         inneriter                  += j_index_end - j_index_start;
 416
 417         /* Outer loop uses 9 flops */
 418     }
 419
 420     /* Increment number of outer iterations */
 421     outeriter        += nri;
 422
 423     /* Update outer/inner flops */
 424
 425     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 426 }
 427 /*
 428  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_double
 429  * Electrostatics interaction: Ewald
 430  * VdW interaction:            LennardJones
 431  * Geometry:                   Particle-Particle
 432  * Calculate force/pot:        Force
 433  */
 434 void
 435 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_double
 436                     (t_nblist                    * gmx_restrict       nlist,
 437                      rvec                        * gmx_restrict          xx,
 438                      rvec                        * gmx_restrict          ff,
 439                      t_forcerec                  * gmx_restrict          fr,
 440                      t_mdatoms                   * gmx_restrict     mdatoms,
 441                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 442                      t_nrnb                      * gmx_restrict        nrnb)
 443 {
 444     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 445      * just 0 for non-waters.
 446      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 447      * jnr indices corresponding to data put in the four positions in the SIMD register.
 448      */
 449     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 450     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 451     int              jnrA,jnrB;
 452     int              j_coord_offsetA,j_coord_offsetB;
 453     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 454     real             rcutoff_scalar;
 455     real             *shiftvec,*fshift,*x,*f;
 456     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 457     int              vdwioffset0;
 458     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 459     int              vdwjidx0A,vdwjidx0B;
 460     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 461     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 462     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 463     real             *charge;
 464     int              nvdwtype;
 465     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 466     int              *vdwtype;
 467     real             *vdwparam;
 468     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 469     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 470     __m128i          ewitab;
 471     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 472     real             *ewtab;
 473     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 474     real             rswitch_scalar,d_scalar;
 475     __m128d          dummy_mask,cutoff_mask;
 476     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 477     __m128d          one     = _mm_set1_pd(1.0);
 478     __m128d          two     = _mm_set1_pd(2.0);
 479     x                = xx[0];
 480     f                = ff[0];
 481
 482     nri              = nlist->nri;
 483     iinr             = nlist->iinr;
 484     jindex           = nlist->jindex;
 485     jjnr             = nlist->jjnr;
 486     shiftidx         = nlist->shift;
 487     gid              = nlist->gid;
 488     shiftvec         = fr->shift_vec[0];
 489     fshift           = fr->fshift[0];
 490     facel            = _mm_set1_pd(fr->epsfac);
 491     charge           = mdatoms->chargeA;
 492     nvdwtype         = fr->ntype;
 493     vdwparam         = fr->nbfp;
 494     vdwtype          = mdatoms->typeA;
 495
 496     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 497     ewtab            = fr->ic->tabq_coul_FDV0;
 498     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 499     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 500
 501     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 502     rcutoff_scalar   = fr->rcoulomb;
 503     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 504     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 505
 506     rswitch_scalar   = fr->rcoulomb_switch;
 507     rswitch          = _mm_set1_pd(rswitch_scalar);
 508     /* Setup switch parameters */
 509     d_scalar         = rcutoff_scalar-rswitch_scalar;
 510     d                = _mm_set1_pd(d_scalar);
 511     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 512     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 513     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 514     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 515     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 516     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 517
 518     /* Avoid stupid compiler warnings */
 519     jnrA = jnrB = 0;
 520     j_coord_offsetA = 0;
 521     j_coord_offsetB = 0;
 522
 523     outeriter        = 0;
 524     inneriter        = 0;
 525
 526     /* Start outer loop over neighborlists */
 527     for(iidx=0; iidx<nri; iidx++)
 528     {
 529         /* Load shift vector for this list */
 530         i_shift_offset   = DIM*shiftidx[iidx];
 531
 532         /* Load limits for loop over neighbors */
 533         j_index_start    = jindex[iidx];
 534         j_index_end      = jindex[iidx+1];
 535
 536         /* Get outer coordinate index */
 537         inr              = iinr[iidx];
 538         i_coord_offset   = DIM*inr;
 539
 540         /* Load i particle coords and add shift vector */
 541         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 542
 543         fix0             = _mm_setzero_pd();
 544         fiy0             = _mm_setzero_pd();
 545         fiz0             = _mm_setzero_pd();
 546
 547         /* Load parameters for i particles */
 548         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 549         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 550
 551         /* Start inner kernel loop */
 552         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 553         {
 554
 555             /* Get j neighbor index, and coordinate index */
 556             jnrA             = jjnr[jidx];
 557             jnrB             = jjnr[jidx+1];
 558             j_coord_offsetA  = DIM*jnrA;
 559             j_coord_offsetB  = DIM*jnrB;
 560
 561             /* load j atom coordinates */
 562             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 563                                               &jx0,&jy0,&jz0);
 564
 565             /* Calculate displacement vector */
 566             dx00             = _mm_sub_pd(ix0,jx0);
 567             dy00             = _mm_sub_pd(iy0,jy0);
 568             dz00             = _mm_sub_pd(iz0,jz0);
 569
 570             /* Calculate squared distance and things based on it */
 571             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 572
 573             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 574
 575             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 576
 577             /* Load parameters for j particles */
 578             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 579             vdwjidx0A        = 2*vdwtype[jnrA+0];
 580             vdwjidx0B        = 2*vdwtype[jnrB+0];
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             if (gmx_mm_any_lt(rsq00,rcutoff2))
 587             {
 588
 589             r00              = _mm_mul_pd(rsq00,rinv00);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq00             = _mm_mul_pd(iq0,jq0);
 593             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 594                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 595
 596             /* EWALD ELECTROSTATICS */
 597
 598             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 599             ewrt             = _mm_mul_pd(r00,ewtabscale);
 600             ewitab           = _mm_cvttpd_epi32(ewrt);
 601             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 602             ewitab           = _mm_slli_epi32(ewitab,2);
 603             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 604             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 605             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 606             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 607             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 608             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 609             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 610             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 611             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 612             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 613
 614             /* LENNARD-JONES DISPERSION/REPULSION */
 615
 616             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 617             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 618             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 619             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 620             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 621
 622             d                = _mm_sub_pd(r00,rswitch);
 623             d                = _mm_max_pd(d,_mm_setzero_pd());
 624             d2               = _mm_mul_pd(d,d);
 625             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 626
 627             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 628
 629             /* Evaluate switch function */
 630             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 631             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 632             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 633             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 634
 635             fscal            = _mm_add_pd(felec,fvdw);
 636
 637             fscal            = _mm_and_pd(fscal,cutoff_mask);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm_mul_pd(fscal,dx00);
 641             ty               = _mm_mul_pd(fscal,dy00);
 642             tz               = _mm_mul_pd(fscal,dz00);
 643
 644             /* Update vectorial force */
 645             fix0             = _mm_add_pd(fix0,tx);
 646             fiy0             = _mm_add_pd(fiy0,ty);
 647             fiz0             = _mm_add_pd(fiz0,tz);
 648
 649             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 650
 651             }
 652
 653             /* Inner loop uses 77 flops */
 654         }
 655
 656         if(jidx<j_index_end)
 657         {
 658
 659             jnrA             = jjnr[jidx];
 660             j_coord_offsetA  = DIM*jnrA;
 661
 662             /* load j atom coordinates */
 663             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 664                                               &jx0,&jy0,&jz0);
 665
 666             /* Calculate displacement vector */
 667             dx00             = _mm_sub_pd(ix0,jx0);
 668             dy00             = _mm_sub_pd(iy0,jy0);
 669             dz00             = _mm_sub_pd(iz0,jz0);
 670
 671             /* Calculate squared distance and things based on it */
 672             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 673
 674             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 675
 676             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 677
 678             /* Load parameters for j particles */
 679             jq0              = _mm_load_sd(charge+jnrA+0);
 680             vdwjidx0A        = 2*vdwtype[jnrA+0];
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             if (gmx_mm_any_lt(rsq00,rcutoff2))
 687             {
 688
 689             r00              = _mm_mul_pd(rsq00,rinv00);
 690
 691             /* Compute parameters for interactions between i and j atoms */
 692             qq00             = _mm_mul_pd(iq0,jq0);
 693             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 694
 695             /* EWALD ELECTROSTATICS */
 696
 697             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 698             ewrt             = _mm_mul_pd(r00,ewtabscale);
 699             ewitab           = _mm_cvttpd_epi32(ewrt);
 700             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 701             ewitab           = _mm_slli_epi32(ewitab,2);
 702             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 703             ewtabD           = _mm_setzero_pd();
 704             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 705             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 706             ewtabFn          = _mm_setzero_pd();
 707             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 708             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 709             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 710             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 711             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 712
 713             /* LENNARD-JONES DISPERSION/REPULSION */
 714
 715             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 716             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 717             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 718             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 719             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 720
 721             d                = _mm_sub_pd(r00,rswitch);
 722             d                = _mm_max_pd(d,_mm_setzero_pd());
 723             d2               = _mm_mul_pd(d,d);
 724             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 725
 726             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 727
 728             /* Evaluate switch function */
 729             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 730             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 731             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 732             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 733
 734             fscal            = _mm_add_pd(felec,fvdw);
 735
 736             fscal            = _mm_and_pd(fscal,cutoff_mask);
 737
 738             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm_mul_pd(fscal,dx00);
 742             ty               = _mm_mul_pd(fscal,dy00);
 743             tz               = _mm_mul_pd(fscal,dz00);
 744
 745             /* Update vectorial force */
 746             fix0             = _mm_add_pd(fix0,tx);
 747             fiy0             = _mm_add_pd(fiy0,ty);
 748             fiz0             = _mm_add_pd(fiz0,tz);
 749
 750             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 751
 752             }
 753
 754             /* Inner loop uses 77 flops */
 755         }
 756
 757         /* End of innermost loop */
 758
 759         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 760                                               f+i_coord_offset,fshift+i_shift_offset);
 761
 762         /* Increment number of inner iterations */
 763         inneriter                  += j_index_end - j_index_start;
 764
 765         /* Outer loop uses 7 flops */
 766     }
 767
 768     /* Increment number of outer iterations */
 769     outeriter        += nri;
 770
 771     /* Update outer/inner flops */
 772
 773     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*77);
 774 }