src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->rcoulomb_switch;
 130     rswitch          = _mm_set1_pd(rswitch_scalar);
 131     /* Setup switch parameters */
 132     d_scalar         = rcutoff_scalar-rswitch_scalar;
 133     d                = _mm_set1_pd(d_scalar);
 134     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 135     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 138     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_pd();
 176         vvdwsum          = _mm_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_pd(ix0,jx0);
 194             dy00             = _mm_sub_pd(iy0,jy0);
 195             dz00             = _mm_sub_pd(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201
 202             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_mm_any_lt(rsq00,rcutoff2))
 214             {
 215
 216             r00              = _mm_mul_pd(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             /* EWALD ELECTROSTATICS */
 224
 225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 226             ewrt             = _mm_mul_pd(r00,ewtabscale);
 227             ewitab           = _mm_cvttpd_epi32(ewrt);
 228             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 229             ewitab           = _mm_slli_epi32(ewitab,2);
 230             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 231             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 232             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 233             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 234             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 235             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 236             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 237             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 238             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 239             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 245             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 246             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 247             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 248
 249             d                = _mm_sub_pd(r00,rswitch);
 250             d                = _mm_max_pd(d,_mm_setzero_pd());
 251             d2               = _mm_mul_pd(d,d);
 252             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)))))));
 253
 254             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 255
 256             /* Evaluate switch function */
 257             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 258             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 259             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 260             velec            = _mm_mul_pd(velec,sw);
 261             vvdw             = _mm_mul_pd(vvdw,sw);
 262             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velec            = _mm_and_pd(velec,cutoff_mask);
 266             velecsum         = _mm_add_pd(velecsum,velec);
 267             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 268             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 269
 270             fscal            = _mm_add_pd(felec,fvdw);
 271
 272             fscal            = _mm_and_pd(fscal,cutoff_mask);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_pd(fscal,dx00);
 276             ty               = _mm_mul_pd(fscal,dy00);
 277             tz               = _mm_mul_pd(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_pd(fix0,tx);
 281             fiy0             = _mm_add_pd(fiy0,ty);
 282             fiz0             = _mm_add_pd(fiz0,tz);
 283
 284             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 285
 286             }
 287
 288             /* Inner loop uses 83 flops */
 289         }
 290
 291         if(jidx<j_index_end)
 292         {
 293
 294             jnrA             = jjnr[jidx];
 295             j_coord_offsetA  = DIM*jnrA;
 296
 297             /* load j atom coordinates */
 298             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 299                                               &jx0,&jy0,&jz0);
 300
 301             /* Calculate displacement vector */
 302             dx00             = _mm_sub_pd(ix0,jx0);
 303             dy00             = _mm_sub_pd(iy0,jy0);
 304             dz00             = _mm_sub_pd(iz0,jz0);
 305
 306             /* Calculate squared distance and things based on it */
 307             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 308
 309             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 310
 311             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 312
 313             /* Load parameters for j particles */
 314             jq0              = _mm_load_sd(charge+jnrA+0);
 315             vdwjidx0A        = 2*vdwtype[jnrA+0];
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm_any_lt(rsq00,rcutoff2))
 322             {
 323
 324             r00              = _mm_mul_pd(rsq00,rinv00);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq00             = _mm_mul_pd(iq0,jq0);
 328             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 329
 330             /* EWALD ELECTROSTATICS */
 331
 332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 333             ewrt             = _mm_mul_pd(r00,ewtabscale);
 334             ewitab           = _mm_cvttpd_epi32(ewrt);
 335             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 336             ewitab           = _mm_slli_epi32(ewitab,2);
 337             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 338             ewtabD           = _mm_setzero_pd();
 339             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 340             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 341             ewtabFn          = _mm_setzero_pd();
 342             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 343             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 344             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 345             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 346             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 347
 348             /* LENNARD-JONES DISPERSION/REPULSION */
 349
 350             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 351             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 352             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 353             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 354             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 355
 356             d                = _mm_sub_pd(r00,rswitch);
 357             d                = _mm_max_pd(d,_mm_setzero_pd());
 358             d2               = _mm_mul_pd(d,d);
 359             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)))))));
 360
 361             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 362
 363             /* Evaluate switch function */
 364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 365             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 366             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 367             velec            = _mm_mul_pd(velec,sw);
 368             vvdw             = _mm_mul_pd(vvdw,sw);
 369             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _mm_and_pd(velec,cutoff_mask);
 373             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 374             velecsum         = _mm_add_pd(velecsum,velec);
 375             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 376             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 377             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 378
 379             fscal            = _mm_add_pd(felec,fvdw);
 380
 381             fscal            = _mm_and_pd(fscal,cutoff_mask);
 382
 383             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_pd(fscal,dx00);
 387             ty               = _mm_mul_pd(fscal,dy00);
 388             tz               = _mm_mul_pd(fscal,dz00);
 389
 390             /* Update vectorial force */
 391             fix0             = _mm_add_pd(fix0,tx);
 392             fiy0             = _mm_add_pd(fiy0,ty);
 393             fiz0             = _mm_add_pd(fiz0,tz);
 394
 395             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 396
 397             }
 398
 399             /* Inner loop uses 83 flops */
 400         }
 401
 402         /* End of innermost loop */
 403
 404         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 405                                               f+i_coord_offset,fshift+i_shift_offset);
 406
 407         ggid                        = gid[iidx];
 408         /* Update potential energies */
 409         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 410         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 411
 412         /* Increment number of inner iterations */
 413         inneriter                  += j_index_end - j_index_start;
 414
 415         /* Outer loop uses 9 flops */
 416     }
 417
 418     /* Increment number of outer iterations */
 419     outeriter        += nri;
 420
 421     /* Update outer/inner flops */
 422
 423     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 424 }
 425 /*
 426  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
 427  * Electrostatics interaction: Ewald
 428  * VdW interaction:            LennardJones
 429  * Geometry:                   Particle-Particle
 430  * Calculate force/pot:        Force
 431  */
 432 void
 433 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
 434                     (t_nblist                    * gmx_restrict       nlist,
 435                      rvec                        * gmx_restrict          xx,
 436                      rvec                        * gmx_restrict          ff,
 437                      t_forcerec                  * gmx_restrict          fr,
 438                      t_mdatoms                   * gmx_restrict     mdatoms,
 439                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 440                      t_nrnb                      * gmx_restrict        nrnb)
 441 {
 442     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 443      * just 0 for non-waters.
 444      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 445      * jnr indices corresponding to data put in the four positions in the SIMD register.
 446      */
 447     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 448     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 449     int              jnrA,jnrB;
 450     int              j_coord_offsetA,j_coord_offsetB;
 451     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 452     real             rcutoff_scalar;
 453     real             *shiftvec,*fshift,*x,*f;
 454     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 455     int              vdwioffset0;
 456     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 457     int              vdwjidx0A,vdwjidx0B;
 458     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 459     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 460     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 461     real             *charge;
 462     int              nvdwtype;
 463     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 464     int              *vdwtype;
 465     real             *vdwparam;
 466     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 467     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 468     __m128i          ewitab;
 469     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 470     real             *ewtab;
 471     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 472     real             rswitch_scalar,d_scalar;
 473     __m128d          dummy_mask,cutoff_mask;
 474     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 475     __m128d          one     = _mm_set1_pd(1.0);
 476     __m128d          two     = _mm_set1_pd(2.0);
 477     x                = xx[0];
 478     f                = ff[0];
 479
 480     nri              = nlist->nri;
 481     iinr             = nlist->iinr;
 482     jindex           = nlist->jindex;
 483     jjnr             = nlist->jjnr;
 484     shiftidx         = nlist->shift;
 485     gid              = nlist->gid;
 486     shiftvec         = fr->shift_vec[0];
 487     fshift           = fr->fshift[0];
 488     facel            = _mm_set1_pd(fr->epsfac);
 489     charge           = mdatoms->chargeA;
 490     nvdwtype         = fr->ntype;
 491     vdwparam         = fr->nbfp;
 492     vdwtype          = mdatoms->typeA;
 493
 494     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 495     ewtab            = fr->ic->tabq_coul_FDV0;
 496     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 497     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 498
 499     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 500     rcutoff_scalar   = fr->rcoulomb;
 501     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 502     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 503
 504     rswitch_scalar   = fr->rcoulomb_switch;
 505     rswitch          = _mm_set1_pd(rswitch_scalar);
 506     /* Setup switch parameters */
 507     d_scalar         = rcutoff_scalar-rswitch_scalar;
 508     d                = _mm_set1_pd(d_scalar);
 509     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 510     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 511     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 512     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 513     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 514     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 515
 516     /* Avoid stupid compiler warnings */
 517     jnrA = jnrB = 0;
 518     j_coord_offsetA = 0;
 519     j_coord_offsetB = 0;
 520
 521     outeriter        = 0;
 522     inneriter        = 0;
 523
 524     /* Start outer loop over neighborlists */
 525     for(iidx=0; iidx<nri; iidx++)
 526     {
 527         /* Load shift vector for this list */
 528         i_shift_offset   = DIM*shiftidx[iidx];
 529
 530         /* Load limits for loop over neighbors */
 531         j_index_start    = jindex[iidx];
 532         j_index_end      = jindex[iidx+1];
 533
 534         /* Get outer coordinate index */
 535         inr              = iinr[iidx];
 536         i_coord_offset   = DIM*inr;
 537
 538         /* Load i particle coords and add shift vector */
 539         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 540
 541         fix0             = _mm_setzero_pd();
 542         fiy0             = _mm_setzero_pd();
 543         fiz0             = _mm_setzero_pd();
 544
 545         /* Load parameters for i particles */
 546         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 547         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 548
 549         /* Start inner kernel loop */
 550         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 551         {
 552
 553             /* Get j neighbor index, and coordinate index */
 554             jnrA             = jjnr[jidx];
 555             jnrB             = jjnr[jidx+1];
 556             j_coord_offsetA  = DIM*jnrA;
 557             j_coord_offsetB  = DIM*jnrB;
 558
 559             /* load j atom coordinates */
 560             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 561                                               &jx0,&jy0,&jz0);
 562
 563             /* Calculate displacement vector */
 564             dx00             = _mm_sub_pd(ix0,jx0);
 565             dy00             = _mm_sub_pd(iy0,jy0);
 566             dz00             = _mm_sub_pd(iz0,jz0);
 567
 568             /* Calculate squared distance and things based on it */
 569             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 570
 571             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 572
 573             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 574
 575             /* Load parameters for j particles */
 576             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 577             vdwjidx0A        = 2*vdwtype[jnrA+0];
 578             vdwjidx0B        = 2*vdwtype[jnrB+0];
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq00,rcutoff2))
 585             {
 586
 587             r00              = _mm_mul_pd(rsq00,rinv00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq00             = _mm_mul_pd(iq0,jq0);
 591             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 592                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_pd(r00,ewtabscale);
 598             ewitab           = _mm_cvttpd_epi32(ewrt);
 599             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 603             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 604             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 605             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 606             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 607             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 608             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 609             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 610             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 611
 612             /* LENNARD-JONES DISPERSION/REPULSION */
 613
 614             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 615             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 616             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 617             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 618             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 619
 620             d                = _mm_sub_pd(r00,rswitch);
 621             d                = _mm_max_pd(d,_mm_setzero_pd());
 622             d2               = _mm_mul_pd(d,d);
 623             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)))))));
 624
 625             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 626
 627             /* Evaluate switch function */
 628             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 629             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 630             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 631             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 632
 633             fscal            = _mm_add_pd(felec,fvdw);
 634
 635             fscal            = _mm_and_pd(fscal,cutoff_mask);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm_mul_pd(fscal,dx00);
 639             ty               = _mm_mul_pd(fscal,dy00);
 640             tz               = _mm_mul_pd(fscal,dz00);
 641
 642             /* Update vectorial force */
 643             fix0             = _mm_add_pd(fix0,tx);
 644             fiy0             = _mm_add_pd(fiy0,ty);
 645             fiz0             = _mm_add_pd(fiz0,tz);
 646
 647             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 648
 649             }
 650
 651             /* Inner loop uses 77 flops */
 652         }
 653
 654         if(jidx<j_index_end)
 655         {
 656
 657             jnrA             = jjnr[jidx];
 658             j_coord_offsetA  = DIM*jnrA;
 659
 660             /* load j atom coordinates */
 661             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 662                                               &jx0,&jy0,&jz0);
 663
 664             /* Calculate displacement vector */
 665             dx00             = _mm_sub_pd(ix0,jx0);
 666             dy00             = _mm_sub_pd(iy0,jy0);
 667             dz00             = _mm_sub_pd(iz0,jz0);
 668
 669             /* Calculate squared distance and things based on it */
 670             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 671
 672             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 673
 674             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 675
 676             /* Load parameters for j particles */
 677             jq0              = _mm_load_sd(charge+jnrA+0);
 678             vdwjidx0A        = 2*vdwtype[jnrA+0];
 679
 680             /**************************
 681              * CALCULATE INTERACTIONS *
 682              **************************/
 683
 684             if (gmx_mm_any_lt(rsq00,rcutoff2))
 685             {
 686
 687             r00              = _mm_mul_pd(rsq00,rinv00);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq00             = _mm_mul_pd(iq0,jq0);
 691             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 692
 693             /* EWALD ELECTROSTATICS */
 694
 695             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 696             ewrt             = _mm_mul_pd(r00,ewtabscale);
 697             ewitab           = _mm_cvttpd_epi32(ewrt);
 698             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 699             ewitab           = _mm_slli_epi32(ewitab,2);
 700             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 701             ewtabD           = _mm_setzero_pd();
 702             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 703             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 704             ewtabFn          = _mm_setzero_pd();
 705             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 706             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 707             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 708             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 709             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 710
 711             /* LENNARD-JONES DISPERSION/REPULSION */
 712
 713             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 714             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 715             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 716             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 717             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 718
 719             d                = _mm_sub_pd(r00,rswitch);
 720             d                = _mm_max_pd(d,_mm_setzero_pd());
 721             d2               = _mm_mul_pd(d,d);
 722             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)))))));
 723
 724             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 725
 726             /* Evaluate switch function */
 727             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 728             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 729             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 730             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 731
 732             fscal            = _mm_add_pd(felec,fvdw);
 733
 734             fscal            = _mm_and_pd(fscal,cutoff_mask);
 735
 736             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 737
 738             /* Calculate temporary vectorial force */
 739             tx               = _mm_mul_pd(fscal,dx00);
 740             ty               = _mm_mul_pd(fscal,dy00);
 741             tz               = _mm_mul_pd(fscal,dz00);
 742
 743             /* Update vectorial force */
 744             fix0             = _mm_add_pd(fix0,tx);
 745             fiy0             = _mm_add_pd(fiy0,ty);
 746             fiz0             = _mm_add_pd(fiz0,tz);
 747
 748             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 749
 750             }
 751
 752             /* Inner loop uses 77 flops */
 753         }
 754
 755         /* End of innermost loop */
 756
 757         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 758                                               f+i_coord_offset,fshift+i_shift_offset);
 759
 760         /* Increment number of inner iterations */
 761         inneriter                  += j_index_end - j_index_start;
 762
 763         /* Outer loop uses 7 flops */
 764     }
 765
 766     /* Increment number of outer iterations */
 767     outeriter        += nri;
 768
 769     /* Update outer/inner flops */
 770
 771     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*77);
 772 }