src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_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          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_pd(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 190
 191             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _mm_mul_pd(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_pd(iq0,jq0);
 209             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _mm_mul_pd(r00,ewtabscale);
 216             ewitab           = _mm_cvttpd_epi32(ewrt);
 217             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 218             ewitab           = _mm_slli_epi32(ewitab,2);
 219             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 220             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 222             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 223             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 225             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 226             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 227             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 228             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 229
 230             /* LENNARD-JONES DISPERSION/REPULSION */
 231
 232             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 233             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 234             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 235             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) ,
 236                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 237             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 238
 239             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velec            = _mm_and_pd(velec,cutoff_mask);
 243             velecsum         = _mm_add_pd(velecsum,velec);
 244             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 245             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 246
 247             fscal            = _mm_add_pd(felec,fvdw);
 248
 249             fscal            = _mm_and_pd(fscal,cutoff_mask);
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm_mul_pd(fscal,dx00);
 253             ty               = _mm_mul_pd(fscal,dy00);
 254             tz               = _mm_mul_pd(fscal,dz00);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm_add_pd(fix0,tx);
 258             fiy0             = _mm_add_pd(fiy0,ty);
 259             fiz0             = _mm_add_pd(fiz0,tz);
 260
 261             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 262
 263             }
 264
 265             /* Inner loop uses 64 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_pd(ix0,jx0);
 280             dy00             = _mm_sub_pd(iy0,jy0);
 281             dz00             = _mm_sub_pd(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 287
 288             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _mm_mul_pd(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_pd(iq0,jq0);
 305             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r00,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm_setzero_pd();
 316             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 317             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 318             ewtabFn          = _mm_setzero_pd();
 319             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 320             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 321             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 322             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 323             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 324
 325             /* LENNARD-JONES DISPERSION/REPULSION */
 326
 327             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 328             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 329             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 330             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) ,
 331                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 332             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 333
 334             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_and_pd(velec,cutoff_mask);
 338             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 341             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 342             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm_add_pd(felec,fvdw);
 345
 346             fscal            = _mm_and_pd(fscal,cutoff_mask);
 347
 348             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm_mul_pd(fscal,dx00);
 352             ty               = _mm_mul_pd(fscal,dy00);
 353             tz               = _mm_mul_pd(fscal,dz00);
 354
 355             /* Update vectorial force */
 356             fix0             = _mm_add_pd(fix0,tx);
 357             fiy0             = _mm_add_pd(fiy0,ty);
 358             fiz0             = _mm_add_pd(fiz0,tz);
 359
 360             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 361
 362             }
 363
 364             /* Inner loop uses 64 flops */
 365         }
 366
 367         /* End of innermost loop */
 368
 369         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 370                                               f+i_coord_offset,fshift+i_shift_offset);
 371
 372         ggid                        = gid[iidx];
 373         /* Update potential energies */
 374         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 375         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 376
 377         /* Increment number of inner iterations */
 378         inneriter                  += j_index_end - j_index_start;
 379
 380         /* Outer loop uses 9 flops */
 381     }
 382
 383     /* Increment number of outer iterations */
 384     outeriter        += nri;
 385
 386     /* Update outer/inner flops */
 387
 388     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
 389 }
 390 /*
 391  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_double
 392  * Electrostatics interaction: Ewald
 393  * VdW interaction:            LennardJones
 394  * Geometry:                   Particle-Particle
 395  * Calculate force/pot:        Force
 396  */
 397 void
 398 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_double
 399                     (t_nblist                    * gmx_restrict       nlist,
 400                      rvec                        * gmx_restrict          xx,
 401                      rvec                        * gmx_restrict          ff,
 402                      t_forcerec                  * gmx_restrict          fr,
 403                      t_mdatoms                   * gmx_restrict     mdatoms,
 404                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 405                      t_nrnb                      * gmx_restrict        nrnb)
 406 {
 407     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 408      * just 0 for non-waters.
 409      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 410      * jnr indices corresponding to data put in the four positions in the SIMD register.
 411      */
 412     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 413     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 414     int              jnrA,jnrB;
 415     int              j_coord_offsetA,j_coord_offsetB;
 416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 417     real             rcutoff_scalar;
 418     real             *shiftvec,*fshift,*x,*f;
 419     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 420     int              vdwioffset0;
 421     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 422     int              vdwjidx0A,vdwjidx0B;
 423     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 424     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 425     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 426     real             *charge;
 427     int              nvdwtype;
 428     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 429     int              *vdwtype;
 430     real             *vdwparam;
 431     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 432     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 433     __m128i          ewitab;
 434     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 435     real             *ewtab;
 436     __m128d          dummy_mask,cutoff_mask;
 437     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 438     __m128d          one     = _mm_set1_pd(1.0);
 439     __m128d          two     = _mm_set1_pd(2.0);
 440     x                = xx[0];
 441     f                = ff[0];
 442
 443     nri              = nlist->nri;
 444     iinr             = nlist->iinr;
 445     jindex           = nlist->jindex;
 446     jjnr             = nlist->jjnr;
 447     shiftidx         = nlist->shift;
 448     gid              = nlist->gid;
 449     shiftvec         = fr->shift_vec[0];
 450     fshift           = fr->fshift[0];
 451     facel            = _mm_set1_pd(fr->epsfac);
 452     charge           = mdatoms->chargeA;
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456
 457     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 458     ewtab            = fr->ic->tabq_coul_F;
 459     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 460     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 461
 462     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 463     rcutoff_scalar   = fr->rcoulomb;
 464     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 465     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 466
 467     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 468     rvdw             = _mm_set1_pd(fr->rvdw);
 469
 470     /* Avoid stupid compiler warnings */
 471     jnrA = jnrB = 0;
 472     j_coord_offsetA = 0;
 473     j_coord_offsetB = 0;
 474
 475     outeriter        = 0;
 476     inneriter        = 0;
 477
 478     /* Start outer loop over neighborlists */
 479     for(iidx=0; iidx<nri; iidx++)
 480     {
 481         /* Load shift vector for this list */
 482         i_shift_offset   = DIM*shiftidx[iidx];
 483
 484         /* Load limits for loop over neighbors */
 485         j_index_start    = jindex[iidx];
 486         j_index_end      = jindex[iidx+1];
 487
 488         /* Get outer coordinate index */
 489         inr              = iinr[iidx];
 490         i_coord_offset   = DIM*inr;
 491
 492         /* Load i particle coords and add shift vector */
 493         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 494
 495         fix0             = _mm_setzero_pd();
 496         fiy0             = _mm_setzero_pd();
 497         fiz0             = _mm_setzero_pd();
 498
 499         /* Load parameters for i particles */
 500         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 501         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 502
 503         /* Start inner kernel loop */
 504         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 505         {
 506
 507             /* Get j neighbor index, and coordinate index */
 508             jnrA             = jjnr[jidx];
 509             jnrB             = jjnr[jidx+1];
 510             j_coord_offsetA  = DIM*jnrA;
 511             j_coord_offsetB  = DIM*jnrB;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm_sub_pd(ix0,jx0);
 519             dy00             = _mm_sub_pd(iy0,jy0);
 520             dz00             = _mm_sub_pd(iz0,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 524
 525             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 526
 527             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 528
 529             /* Load parameters for j particles */
 530             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 531             vdwjidx0A        = 2*vdwtype[jnrA+0];
 532             vdwjidx0B        = 2*vdwtype[jnrB+0];
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq00,rcutoff2))
 539             {
 540
 541             r00              = _mm_mul_pd(rsq00,rinv00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq00             = _mm_mul_pd(iq0,jq0);
 545             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 546                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 551             ewrt             = _mm_mul_pd(r00,ewtabscale);
 552             ewitab           = _mm_cvttpd_epi32(ewrt);
 553             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 554             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 555                                          &ewtabF,&ewtabFn);
 556             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 557             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 558
 559             /* LENNARD-JONES DISPERSION/REPULSION */
 560
 561             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 562             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 563
 564             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 565
 566             fscal            = _mm_add_pd(felec,fvdw);
 567
 568             fscal            = _mm_and_pd(fscal,cutoff_mask);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm_mul_pd(fscal,dx00);
 572             ty               = _mm_mul_pd(fscal,dy00);
 573             tz               = _mm_mul_pd(fscal,dz00);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_add_pd(fix0,tx);
 577             fiy0             = _mm_add_pd(fiy0,ty);
 578             fiz0             = _mm_add_pd(fiz0,tz);
 579
 580             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 581
 582             }
 583
 584             /* Inner loop uses 46 flops */
 585         }
 586
 587         if(jidx<j_index_end)
 588         {
 589
 590             jnrA             = jjnr[jidx];
 591             j_coord_offsetA  = DIM*jnrA;
 592
 593             /* load j atom coordinates */
 594             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 595                                               &jx0,&jy0,&jz0);
 596
 597             /* Calculate displacement vector */
 598             dx00             = _mm_sub_pd(ix0,jx0);
 599             dy00             = _mm_sub_pd(iy0,jy0);
 600             dz00             = _mm_sub_pd(iz0,jz0);
 601
 602             /* Calculate squared distance and things based on it */
 603             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 604
 605             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 606
 607             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 608
 609             /* Load parameters for j particles */
 610             jq0              = _mm_load_sd(charge+jnrA+0);
 611             vdwjidx0A        = 2*vdwtype[jnrA+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq00,rcutoff2))
 618             {
 619
 620             r00              = _mm_mul_pd(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_pd(iq0,jq0);
 624             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_pd(r00,ewtabscale);
 630             ewitab           = _mm_cvttpd_epi32(ewrt);
 631             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 632             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 633             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 634             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 635
 636             /* LENNARD-JONES DISPERSION/REPULSION */
 637
 638             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 639             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 640
 641             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 642
 643             fscal            = _mm_add_pd(felec,fvdw);
 644
 645             fscal            = _mm_and_pd(fscal,cutoff_mask);
 646
 647             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm_mul_pd(fscal,dx00);
 651             ty               = _mm_mul_pd(fscal,dy00);
 652             tz               = _mm_mul_pd(fscal,dz00);
 653
 654             /* Update vectorial force */
 655             fix0             = _mm_add_pd(fix0,tx);
 656             fiy0             = _mm_add_pd(fiy0,ty);
 657             fiz0             = _mm_add_pd(fiz0,tz);
 658
 659             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 660
 661             }
 662
 663             /* Inner loop uses 46 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 683 }