src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_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          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     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 130     rvdw             = _mm_set1_pd(fr->rvdw);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 163         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_pd(ix0,jx0);
 185             dy00             = _mm_sub_pd(iy0,jy0);
 186             dz00             = _mm_sub_pd(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 192
 193             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             if (gmx_mm_any_lt(rsq00,rcutoff2))
 205             {
 206
 207             r00              = _mm_mul_pd(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm_mul_pd(iq0,jq0);
 211             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 212                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 217             ewrt             = _mm_mul_pd(r00,ewtabscale);
 218             ewitab           = _mm_cvttpd_epi32(ewrt);
 219             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 220             ewitab           = _mm_slli_epi32(ewitab,2);
 221             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 222             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 223             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 224             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 225             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 226             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 227             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 228             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 229             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 230             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 231
 232             /* LENNARD-JONES DISPERSION/REPULSION */
 233
 234             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 235             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 236             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 237             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 238                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 239             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 240
 241             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_pd(velec,cutoff_mask);
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_pd(felec,fvdw);
 250
 251             fscal            = _mm_and_pd(fscal,cutoff_mask);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_pd(fscal,dx00);
 255             ty               = _mm_mul_pd(fscal,dy00);
 256             tz               = _mm_mul_pd(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_pd(fix0,tx);
 260             fiy0             = _mm_add_pd(fiy0,ty);
 261             fiz0             = _mm_add_pd(fiz0,tz);
 262
 263             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 264
 265             }
 266
 267             /* Inner loop uses 64 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             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r00,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_setzero_pd();
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_setzero_pd();
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 323             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 324             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 326
 327             /* LENNARD-JONES DISPERSION/REPULSION */
 328
 329             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 330             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 331             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 332             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 333                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 334             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 335
 336             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_pd(velec,cutoff_mask);
 340             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 341             velecsum         = _mm_add_pd(velecsum,velec);
 342             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 343             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 344             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 345
 346             fscal            = _mm_add_pd(felec,fvdw);
 347
 348             fscal            = _mm_and_pd(fscal,cutoff_mask);
 349
 350             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm_mul_pd(fscal,dx00);
 354             ty               = _mm_mul_pd(fscal,dy00);
 355             tz               = _mm_mul_pd(fscal,dz00);
 356
 357             /* Update vectorial force */
 358             fix0             = _mm_add_pd(fix0,tx);
 359             fiy0             = _mm_add_pd(fiy0,ty);
 360             fiz0             = _mm_add_pd(fiz0,tz);
 361
 362             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 363
 364             }
 365
 366             /* Inner loop uses 64 flops */
 367         }
 368
 369         /* End of innermost loop */
 370
 371         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 372                                               f+i_coord_offset,fshift+i_shift_offset);
 373
 374         ggid                        = gid[iidx];
 375         /* Update potential energies */
 376         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 377         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 378
 379         /* Increment number of inner iterations */
 380         inneriter                  += j_index_end - j_index_start;
 381
 382         /* Outer loop uses 9 flops */
 383     }
 384
 385     /* Increment number of outer iterations */
 386     outeriter        += nri;
 387
 388     /* Update outer/inner flops */
 389
 390     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
 391 }
 392 /*
 393  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_double
 394  * Electrostatics interaction: Ewald
 395  * VdW interaction:            LennardJones
 396  * Geometry:                   Particle-Particle
 397  * Calculate force/pot:        Force
 398  */
 399 void
 400 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_double
 401                     (t_nblist                    * gmx_restrict       nlist,
 402                      rvec                        * gmx_restrict          xx,
 403                      rvec                        * gmx_restrict          ff,
 404                      t_forcerec                  * gmx_restrict          fr,
 405                      t_mdatoms                   * gmx_restrict     mdatoms,
 406                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 407                      t_nrnb                      * gmx_restrict        nrnb)
 408 {
 409     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 410      * just 0 for non-waters.
 411      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 412      * jnr indices corresponding to data put in the four positions in the SIMD register.
 413      */
 414     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 415     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 416     int              jnrA,jnrB;
 417     int              j_coord_offsetA,j_coord_offsetB;
 418     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 419     real             rcutoff_scalar;
 420     real             *shiftvec,*fshift,*x,*f;
 421     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B;
 425     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 428     real             *charge;
 429     int              nvdwtype;
 430     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 431     int              *vdwtype;
 432     real             *vdwparam;
 433     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 434     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 435     __m128i          ewitab;
 436     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 437     real             *ewtab;
 438     __m128d          dummy_mask,cutoff_mask;
 439     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 440     __m128d          one     = _mm_set1_pd(1.0);
 441     __m128d          two     = _mm_set1_pd(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     facel            = _mm_set1_pd(fr->epsfac);
 454     charge           = mdatoms->chargeA;
 455     nvdwtype         = fr->ntype;
 456     vdwparam         = fr->nbfp;
 457     vdwtype          = mdatoms->typeA;
 458
 459     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 460     ewtab            = fr->ic->tabq_coul_F;
 461     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 462     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 463
 464     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 465     rcutoff_scalar   = fr->rcoulomb;
 466     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 467     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 468
 469     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 470     rvdw             = _mm_set1_pd(fr->rvdw);
 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             /* EWALD ELECTROSTATICS */
 551
 552             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 553             ewrt             = _mm_mul_pd(r00,ewtabscale);
 554             ewitab           = _mm_cvttpd_epi32(ewrt);
 555             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 556             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 557                                          &ewtabF,&ewtabFn);
 558             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 559             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 560
 561             /* LENNARD-JONES DISPERSION/REPULSION */
 562
 563             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 564             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 565
 566             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 567
 568             fscal            = _mm_add_pd(felec,fvdw);
 569
 570             fscal            = _mm_and_pd(fscal,cutoff_mask);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_pd(fscal,dx00);
 574             ty               = _mm_mul_pd(fscal,dy00);
 575             tz               = _mm_mul_pd(fscal,dz00);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_add_pd(fix0,tx);
 579             fiy0             = _mm_add_pd(fiy0,ty);
 580             fiz0             = _mm_add_pd(fiz0,tz);
 581
 582             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 583
 584             }
 585
 586             /* Inner loop uses 46 flops */
 587         }
 588
 589         if(jidx<j_index_end)
 590         {
 591
 592             jnrA             = jjnr[jidx];
 593             j_coord_offsetA  = DIM*jnrA;
 594
 595             /* load j atom coordinates */
 596             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 597                                               &jx0,&jy0,&jz0);
 598
 599             /* Calculate displacement vector */
 600             dx00             = _mm_sub_pd(ix0,jx0);
 601             dy00             = _mm_sub_pd(iy0,jy0);
 602             dz00             = _mm_sub_pd(iz0,jz0);
 603
 604             /* Calculate squared distance and things based on it */
 605             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 606
 607             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 608
 609             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 610
 611             /* Load parameters for j particles */
 612             jq0              = _mm_load_sd(charge+jnrA+0);
 613             vdwjidx0A        = 2*vdwtype[jnrA+0];
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq00,rcutoff2))
 620             {
 621
 622             r00              = _mm_mul_pd(rsq00,rinv00);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq00             = _mm_mul_pd(iq0,jq0);
 626             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 631             ewrt             = _mm_mul_pd(r00,ewtabscale);
 632             ewitab           = _mm_cvttpd_epi32(ewrt);
 633             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 634             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 635             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 636             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 637
 638             /* LENNARD-JONES DISPERSION/REPULSION */
 639
 640             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 641             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 642
 643             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 644
 645             fscal            = _mm_add_pd(felec,fvdw);
 646
 647             fscal            = _mm_and_pd(fscal,cutoff_mask);
 648
 649             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm_mul_pd(fscal,dx00);
 653             ty               = _mm_mul_pd(fscal,dy00);
 654             tz               = _mm_mul_pd(fscal,dz00);
 655
 656             /* Update vectorial force */
 657             fix0             = _mm_add_pd(fix0,tx);
 658             fiy0             = _mm_add_pd(fiy0,ty);
 659             fiz0             = _mm_add_pd(fiz0,tz);
 660
 661             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 662
 663             }
 664
 665             /* Inner loop uses 46 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 671                                               f+i_coord_offset,fshift+i_shift_offset);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 7 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 685 }