src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_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_ElecEw_VdwLJ_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     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 155         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_pd();
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_pd(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_pd(iq0,jq0);
 200             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 201                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 202
 203             /* EWALD ELECTROSTATICS */
 204
 205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 206             ewrt             = _mm_mul_pd(r00,ewtabscale);
 207             ewitab           = _mm_cvttpd_epi32(ewrt);
 208             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 209             ewitab           = _mm_slli_epi32(ewitab,2);
 210             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 211             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 212             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 213             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 214             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 215             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 216             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 217             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 218             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 219             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 220
 221             /* LENNARD-JONES DISPERSION/REPULSION */
 222
 223             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 224             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 225             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 226             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 227             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velecsum         = _mm_add_pd(velecsum,velec);
 231             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 232
 233             fscal            = _mm_add_pd(felec,fvdw);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm_mul_pd(fscal,dx00);
 237             ty               = _mm_mul_pd(fscal,dy00);
 238             tz               = _mm_mul_pd(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_add_pd(fix0,tx);
 242             fiy0             = _mm_add_pd(fiy0,ty);
 243             fiz0             = _mm_add_pd(fiz0,tz);
 244
 245             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 246
 247             /* Inner loop uses 53 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = _mm_load_sd(charge+jnrA+0);
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r00              = _mm_mul_pd(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq00             = _mm_mul_pd(iq0,jq0);
 284             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 285
 286             /* EWALD ELECTROSTATICS */
 287
 288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 289             ewrt             = _mm_mul_pd(r00,ewtabscale);
 290             ewitab           = _mm_cvttpd_epi32(ewrt);
 291             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 292             ewitab           = _mm_slli_epi32(ewitab,2);
 293             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 294             ewtabD           = _mm_setzero_pd();
 295             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 296             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 297             ewtabFn          = _mm_setzero_pd();
 298             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 299             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 300             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 301             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 302             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 303
 304             /* LENNARD-JONES DISPERSION/REPULSION */
 305
 306             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 307             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 308             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 309             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 310             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 316             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 317
 318             fscal            = _mm_add_pd(felec,fvdw);
 319
 320             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_pd(fscal,dx00);
 324             ty               = _mm_mul_pd(fscal,dy00);
 325             tz               = _mm_mul_pd(fscal,dz00);
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_add_pd(fix0,tx);
 329             fiy0             = _mm_add_pd(fiy0,ty);
 330             fiz0             = _mm_add_pd(fiz0,tz);
 331
 332             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 333
 334             /* Inner loop uses 53 flops */
 335         }
 336
 337         /* End of innermost loop */
 338
 339         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 340                                               f+i_coord_offset,fshift+i_shift_offset);
 341
 342         ggid                        = gid[iidx];
 343         /* Update potential energies */
 344         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 345         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 346
 347         /* Increment number of inner iterations */
 348         inneriter                  += j_index_end - j_index_start;
 349
 350         /* Outer loop uses 9 flops */
 351     }
 352
 353     /* Increment number of outer iterations */
 354     outeriter        += nri;
 355
 356     /* Update outer/inner flops */
 357
 358     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*53);
 359 }
 360 /*
 361  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_double
 362  * Electrostatics interaction: Ewald
 363  * VdW interaction:            LennardJones
 364  * Geometry:                   Particle-Particle
 365  * Calculate force/pot:        Force
 366  */
 367 void
 368 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_double
 369                     (t_nblist                    * gmx_restrict       nlist,
 370                      rvec                        * gmx_restrict          xx,
 371                      rvec                        * gmx_restrict          ff,
 372                      t_forcerec                  * gmx_restrict          fr,
 373                      t_mdatoms                   * gmx_restrict     mdatoms,
 374                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 375                      t_nrnb                      * gmx_restrict        nrnb)
 376 {
 377     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 378      * just 0 for non-waters.
 379      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 380      * jnr indices corresponding to data put in the four positions in the SIMD register.
 381      */
 382     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 383     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 384     int              jnrA,jnrB;
 385     int              j_coord_offsetA,j_coord_offsetB;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 390     int              vdwioffset0;
 391     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B;
 393     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 396     real             *charge;
 397     int              nvdwtype;
 398     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 402     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 403     __m128i          ewitab;
 404     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 405     real             *ewtab;
 406     __m128d          dummy_mask,cutoff_mask;
 407     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 408     __m128d          one     = _mm_set1_pd(1.0);
 409     __m128d          two     = _mm_set1_pd(2.0);
 410     x                = xx[0];
 411     f                = ff[0];
 412
 413     nri              = nlist->nri;
 414     iinr             = nlist->iinr;
 415     jindex           = nlist->jindex;
 416     jjnr             = nlist->jjnr;
 417     shiftidx         = nlist->shift;
 418     gid              = nlist->gid;
 419     shiftvec         = fr->shift_vec[0];
 420     fshift           = fr->fshift[0];
 421     facel            = _mm_set1_pd(fr->epsfac);
 422     charge           = mdatoms->chargeA;
 423     nvdwtype         = fr->ntype;
 424     vdwparam         = fr->nbfp;
 425     vdwtype          = mdatoms->typeA;
 426
 427     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 428     ewtab            = fr->ic->tabq_coul_F;
 429     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 430     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 431
 432     /* Avoid stupid compiler warnings */
 433     jnrA = jnrB = 0;
 434     j_coord_offsetA = 0;
 435     j_coord_offsetB = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_pd();
 458         fiy0             = _mm_setzero_pd();
 459         fiz0             = _mm_setzero_pd();
 460
 461         /* Load parameters for i particles */
 462         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 463         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               &jx0,&jy0,&jz0);
 478
 479             /* Calculate displacement vector */
 480             dx00             = _mm_sub_pd(ix0,jx0);
 481             dy00             = _mm_sub_pd(iy0,jy0);
 482             dz00             = _mm_sub_pd(iz0,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 486
 487             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 488
 489             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 493             vdwjidx0A        = 2*vdwtype[jnrA+0];
 494             vdwjidx0B        = 2*vdwtype[jnrB+0];
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_pd(rsq00,rinv00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm_mul_pd(iq0,jq0);
 504             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 510             ewrt             = _mm_mul_pd(r00,ewtabscale);
 511             ewitab           = _mm_cvttpd_epi32(ewrt);
 512             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 513             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 514                                          &ewtabF,&ewtabFn);
 515             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 516             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 517
 518             /* LENNARD-JONES DISPERSION/REPULSION */
 519
 520             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 521             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 522
 523             fscal            = _mm_add_pd(felec,fvdw);
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm_mul_pd(fscal,dx00);
 527             ty               = _mm_mul_pd(fscal,dy00);
 528             tz               = _mm_mul_pd(fscal,dz00);
 529
 530             /* Update vectorial force */
 531             fix0             = _mm_add_pd(fix0,tx);
 532             fiy0             = _mm_add_pd(fiy0,ty);
 533             fiz0             = _mm_add_pd(fiz0,tz);
 534
 535             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 536
 537             /* Inner loop uses 43 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             jnrA             = jjnr[jidx];
 544             j_coord_offsetA  = DIM*jnrA;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 548                                               &jx0,&jy0,&jz0);
 549
 550             /* Calculate displacement vector */
 551             dx00             = _mm_sub_pd(ix0,jx0);
 552             dy00             = _mm_sub_pd(iy0,jy0);
 553             dz00             = _mm_sub_pd(iz0,jz0);
 554
 555             /* Calculate squared distance and things based on it */
 556             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 557
 558             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 559
 560             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 561
 562             /* Load parameters for j particles */
 563             jq0              = _mm_load_sd(charge+jnrA+0);
 564             vdwjidx0A        = 2*vdwtype[jnrA+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r00              = _mm_mul_pd(rsq00,rinv00);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq00             = _mm_mul_pd(iq0,jq0);
 574             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 575
 576             /* EWALD ELECTROSTATICS */
 577
 578             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 579             ewrt             = _mm_mul_pd(r00,ewtabscale);
 580             ewitab           = _mm_cvttpd_epi32(ewrt);
 581             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 582             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 583             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 584             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 589             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 590
 591             fscal            = _mm_add_pd(felec,fvdw);
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx00);
 597             ty               = _mm_mul_pd(fscal,dy00);
 598             tz               = _mm_mul_pd(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_pd(fix0,tx);
 602             fiy0             = _mm_add_pd(fiy0,ty);
 603             fiz0             = _mm_add_pd(fiz0,tz);
 604
 605             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 606
 607             /* Inner loop uses 43 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         /* Increment number of inner iterations */
 616         inneriter                  += j_index_end - j_index_start;
 617
 618         /* Outer loop uses 7 flops */
 619     }
 620
 621     /* Increment number of outer iterations */
 622     outeriter        += nri;
 623
 624     /* Update outer/inner flops */
 625
 626     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*43);
 627 }