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) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->ic->rcoulomb;
 123     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 127     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = sse41_invsqrt_d(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             r00              = _mm_mul_pd(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_pd(iq0,jq0);
 208             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 210
 211             /* EWALD ELECTROSTATICS */
 212
 213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 214             ewrt             = _mm_mul_pd(r00,ewtabscale);
 215             ewitab           = _mm_cvttpd_epi32(ewrt);
 216             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 217             ewitab           = _mm_slli_epi32(ewitab,2);
 218             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 219             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 220             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 221             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 222             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 223             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 224             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 225             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 226             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 227             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 228
 229             /* LENNARD-JONES DISPERSION/REPULSION */
 230
 231             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 232             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 233             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 234             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) ,
 235                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 236             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 237
 238             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velec            = _mm_and_pd(velec,cutoff_mask);
 242             velecsum         = _mm_add_pd(velecsum,velec);
 243             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 244             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 245
 246             fscal            = _mm_add_pd(felec,fvdw);
 247
 248             fscal            = _mm_and_pd(fscal,cutoff_mask);
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_pd(fscal,dx00);
 252             ty               = _mm_mul_pd(fscal,dy00);
 253             tz               = _mm_mul_pd(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_add_pd(fix0,tx);
 257             fiy0             = _mm_add_pd(fiy0,ty);
 258             fiz0             = _mm_add_pd(fiz0,tz);
 259
 260             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 261
 262             }
 263
 264             /* Inner loop uses 64 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             jnrA             = jjnr[jidx];
 271             j_coord_offsetA  = DIM*jnrA;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_pd(ix0,jx0);
 279             dy00             = _mm_sub_pd(iy0,jy0);
 280             dz00             = _mm_sub_pd(iz0,jz0);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 284
 285             rinv00           = sse41_invsqrt_d(rsq00);
 286
 287             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 288
 289             /* Load parameters for j particles */
 290             jq0              = _mm_load_sd(charge+jnrA+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             if (gmx_mm_any_lt(rsq00,rcutoff2))
 298             {
 299
 300             r00              = _mm_mul_pd(rsq00,rinv00);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq00             = _mm_mul_pd(iq0,jq0);
 304             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_pd(r00,ewtabscale);
 310             ewitab           = _mm_cvttpd_epi32(ewrt);
 311             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm_setzero_pd();
 315             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 316             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 317             ewtabFn          = _mm_setzero_pd();
 318             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 319             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 320             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 321             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 322             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 328             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 329             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) ,
 330                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 331             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 332
 333             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_pd(velec,cutoff_mask);
 337             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 340             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 341             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 342
 343             fscal            = _mm_add_pd(felec,fvdw);
 344
 345             fscal            = _mm_and_pd(fscal,cutoff_mask);
 346
 347             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_pd(fscal,dx00);
 351             ty               = _mm_mul_pd(fscal,dy00);
 352             tz               = _mm_mul_pd(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm_add_pd(fix0,tx);
 356             fiy0             = _mm_add_pd(fiy0,ty);
 357             fiz0             = _mm_add_pd(fiz0,tz);
 358
 359             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 360
 361             }
 362
 363             /* Inner loop uses 64 flops */
 364         }
 365
 366         /* End of innermost loop */
 367
 368         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 369                                               f+i_coord_offset,fshift+i_shift_offset);
 370
 371         ggid                        = gid[iidx];
 372         /* Update potential energies */
 373         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 374         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 9 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_double
 391  * Electrostatics interaction: Ewald
 392  * VdW interaction:            LennardJones
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_double
 398                     (t_nblist                    * gmx_restrict       nlist,
 399                      rvec                        * gmx_restrict          xx,
 400                      rvec                        * gmx_restrict          ff,
 401                      struct t_forcerec           * gmx_restrict          fr,
 402                      t_mdatoms                   * gmx_restrict     mdatoms,
 403                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 404                      t_nrnb                      * gmx_restrict        nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB;
 414     int              j_coord_offsetA,j_coord_offsetB;
 415     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 416     real             rcutoff_scalar;
 417     real             *shiftvec,*fshift,*x,*f;
 418     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B;
 422     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     int              nvdwtype;
 427     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 428     int              *vdwtype;
 429     real             *vdwparam;
 430     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 431     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 432     __m128i          ewitab;
 433     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 434     real             *ewtab;
 435     __m128d          dummy_mask,cutoff_mask;
 436     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 437     __m128d          one     = _mm_set1_pd(1.0);
 438     __m128d          two     = _mm_set1_pd(2.0);
 439     x                = xx[0];
 440     f                = ff[0];
 441
 442     nri              = nlist->nri;
 443     iinr             = nlist->iinr;
 444     jindex           = nlist->jindex;
 445     jjnr             = nlist->jjnr;
 446     shiftidx         = nlist->shift;
 447     gid              = nlist->gid;
 448     shiftvec         = fr->shift_vec[0];
 449     fshift           = fr->fshift[0];
 450     facel            = _mm_set1_pd(fr->ic->epsfac);
 451     charge           = mdatoms->chargeA;
 452     nvdwtype         = fr->ntype;
 453     vdwparam         = fr->nbfp;
 454     vdwtype          = mdatoms->typeA;
 455
 456     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 457     ewtab            = fr->ic->tabq_coul_F;
 458     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 459     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 460
 461     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 462     rcutoff_scalar   = fr->ic->rcoulomb;
 463     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 464     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 465
 466     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 467     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 468
 469     /* Avoid stupid compiler warnings */
 470     jnrA = jnrB = 0;
 471     j_coord_offsetA = 0;
 472     j_coord_offsetB = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 493
 494         fix0             = _mm_setzero_pd();
 495         fiy0             = _mm_setzero_pd();
 496         fiz0             = _mm_setzero_pd();
 497
 498         /* Load parameters for i particles */
 499         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 500         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 501
 502         /* Start inner kernel loop */
 503         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 504         {
 505
 506             /* Get j neighbor index, and coordinate index */
 507             jnrA             = jjnr[jidx];
 508             jnrB             = jjnr[jidx+1];
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511
 512             /* load j atom coordinates */
 513             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _mm_sub_pd(ix0,jx0);
 518             dy00             = _mm_sub_pd(iy0,jy0);
 519             dz00             = _mm_sub_pd(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 523
 524             rinv00           = sse41_invsqrt_d(rsq00);
 525
 526             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 527
 528             /* Load parameters for j particles */
 529             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_mm_any_lt(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _mm_mul_pd(rsq00,rinv00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq00             = _mm_mul_pd(iq0,jq0);
 544             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 545                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 546
 547             /* EWALD ELECTROSTATICS */
 548
 549             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 550             ewrt             = _mm_mul_pd(r00,ewtabscale);
 551             ewitab           = _mm_cvttpd_epi32(ewrt);
 552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 553             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 554                                          &ewtabF,&ewtabFn);
 555             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 556             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 557
 558             /* LENNARD-JONES DISPERSION/REPULSION */
 559
 560             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 561             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 562
 563             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 564
 565             fscal            = _mm_add_pd(felec,fvdw);
 566
 567             fscal            = _mm_and_pd(fscal,cutoff_mask);
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm_mul_pd(fscal,dx00);
 571             ty               = _mm_mul_pd(fscal,dy00);
 572             tz               = _mm_mul_pd(fscal,dz00);
 573
 574             /* Update vectorial force */
 575             fix0             = _mm_add_pd(fix0,tx);
 576             fiy0             = _mm_add_pd(fiy0,ty);
 577             fiz0             = _mm_add_pd(fiz0,tz);
 578
 579             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 580
 581             }
 582
 583             /* Inner loop uses 46 flops */
 584         }
 585
 586         if(jidx<j_index_end)
 587         {
 588
 589             jnrA             = jjnr[jidx];
 590             j_coord_offsetA  = DIM*jnrA;
 591
 592             /* load j atom coordinates */
 593             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 594                                               &jx0,&jy0,&jz0);
 595
 596             /* Calculate displacement vector */
 597             dx00             = _mm_sub_pd(ix0,jx0);
 598             dy00             = _mm_sub_pd(iy0,jy0);
 599             dz00             = _mm_sub_pd(iz0,jz0);
 600
 601             /* Calculate squared distance and things based on it */
 602             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 603
 604             rinv00           = sse41_invsqrt_d(rsq00);
 605
 606             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 607
 608             /* Load parameters for j particles */
 609             jq0              = _mm_load_sd(charge+jnrA+0);
 610             vdwjidx0A        = 2*vdwtype[jnrA+0];
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             if (gmx_mm_any_lt(rsq00,rcutoff2))
 617             {
 618
 619             r00              = _mm_mul_pd(rsq00,rinv00);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq00             = _mm_mul_pd(iq0,jq0);
 623             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_pd(r00,ewtabscale);
 629             ewitab           = _mm_cvttpd_epi32(ewrt);
 630             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 631             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 632             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 633             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 634
 635             /* LENNARD-JONES DISPERSION/REPULSION */
 636
 637             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 638             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 639
 640             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 641
 642             fscal            = _mm_add_pd(felec,fvdw);
 643
 644             fscal            = _mm_and_pd(fscal,cutoff_mask);
 645
 646             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm_mul_pd(fscal,dx00);
 650             ty               = _mm_mul_pd(fscal,dy00);
 651             tz               = _mm_mul_pd(fscal,dz00);
 652
 653             /* Update vectorial force */
 654             fix0             = _mm_add_pd(fix0,tx);
 655             fiy0             = _mm_add_pd(fiy0,ty);
 656             fiz0             = _mm_add_pd(fiz0,tz);
 657
 658             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 659
 660             }
 661
 662             /* Inner loop uses 46 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 668                                               f+i_coord_offset,fshift+i_shift_offset);
 669
 670         /* Increment number of inner iterations */
 671         inneriter                  += j_index_end - j_index_start;
 672
 673         /* Outer loop uses 7 flops */
 674     }
 675
 676     /* Increment number of outer iterations */
 677     outeriter        += nri;
 678
 679     /* Update outer/inner flops */
 680
 681     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 682 }