src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJ_GeomP1P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecEw_VdwLJ_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_ElecEw_VdwLJ_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     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_pd();
 156         vvdwsum          = _mm_setzero_pd();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_pd(ix0,jx0);
 174             dy00             = _mm_sub_pd(iy0,jy0);
 175             dz00             = _mm_sub_pd(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 179
 180             rinv00           = sse41_invsqrt_d(rsq00);
 181
 182             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             r00              = _mm_mul_pd(rsq00,rinv00);
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             qq00             = _mm_mul_pd(iq0,jq0);
 197             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 198                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 199
 200             /* EWALD ELECTROSTATICS */
 201
 202             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 203             ewrt             = _mm_mul_pd(r00,ewtabscale);
 204             ewitab           = _mm_cvttpd_epi32(ewrt);
 205             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 206             ewitab           = _mm_slli_epi32(ewitab,2);
 207             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 208             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 209             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 210             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 211             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 212             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 213             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 214             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 215             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 216             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 217
 218             /* LENNARD-JONES DISPERSION/REPULSION */
 219
 220             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 221             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 222             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 223             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 224             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm_add_pd(velecsum,velec);
 228             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = _mm_add_pd(felec,fvdw);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_pd(fscal,dx00);
 234             ty               = _mm_mul_pd(fscal,dy00);
 235             tz               = _mm_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_pd(fix0,tx);
 239             fiy0             = _mm_add_pd(fiy0,ty);
 240             fiz0             = _mm_add_pd(fiz0,tz);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 243
 244             /* Inner loop uses 53 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_pd(ix0,jx0);
 259             dy00             = _mm_sub_pd(iy0,jy0);
 260             dz00             = _mm_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = sse41_invsqrt_d(rsq00);
 266
 267             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 268
 269             /* Load parameters for j particles */
 270             jq0              = _mm_load_sd(charge+jnrA+0);
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r00              = _mm_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm_mul_pd(iq0,jq0);
 281             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 286             ewrt             = _mm_mul_pd(r00,ewtabscale);
 287             ewitab           = _mm_cvttpd_epi32(ewrt);
 288             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 289             ewitab           = _mm_slli_epi32(ewitab,2);
 290             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 291             ewtabD           = _mm_setzero_pd();
 292             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 293             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 294             ewtabFn          = _mm_setzero_pd();
 295             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 296             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 297             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 298             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 299             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 300
 301             /* LENNARD-JONES DISPERSION/REPULSION */
 302
 303             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 304             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 305             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 306             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 307             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 311             velecsum         = _mm_add_pd(velecsum,velec);
 312             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 313             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 314
 315             fscal            = _mm_add_pd(felec,fvdw);
 316
 317             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_pd(fscal,dx00);
 321             ty               = _mm_mul_pd(fscal,dy00);
 322             tz               = _mm_mul_pd(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm_add_pd(fix0,tx);
 326             fiy0             = _mm_add_pd(fiy0,ty);
 327             fiz0             = _mm_add_pd(fiz0,tz);
 328
 329             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 330
 331             /* Inner loop uses 53 flops */
 332         }
 333
 334         /* End of innermost loop */
 335
 336         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 337                                               f+i_coord_offset,fshift+i_shift_offset);
 338
 339         ggid                        = gid[iidx];
 340         /* Update potential energies */
 341         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 342         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 343
 344         /* Increment number of inner iterations */
 345         inneriter                  += j_index_end - j_index_start;
 346
 347         /* Outer loop uses 9 flops */
 348     }
 349
 350     /* Increment number of outer iterations */
 351     outeriter        += nri;
 352
 353     /* Update outer/inner flops */
 354
 355     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*53);
 356 }
 357 /*
 358  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_double
 359  * Electrostatics interaction: Ewald
 360  * VdW interaction:            LennardJones
 361  * Geometry:                   Particle-Particle
 362  * Calculate force/pot:        Force
 363  */
 364 void
 365 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_double
 366                     (t_nblist                    * gmx_restrict       nlist,
 367                      rvec                        * gmx_restrict          xx,
 368                      rvec                        * gmx_restrict          ff,
 369                      struct t_forcerec           * gmx_restrict          fr,
 370                      t_mdatoms                   * gmx_restrict     mdatoms,
 371                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 372                      t_nrnb                      * gmx_restrict        nrnb)
 373 {
 374     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 375      * just 0 for non-waters.
 376      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 377      * jnr indices corresponding to data put in the four positions in the SIMD register.
 378      */
 379     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 380     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 381     int              jnrA,jnrB;
 382     int              j_coord_offsetA,j_coord_offsetB;
 383     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 384     real             rcutoff_scalar;
 385     real             *shiftvec,*fshift,*x,*f;
 386     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 387     int              vdwioffset0;
 388     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 389     int              vdwjidx0A,vdwjidx0B;
 390     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 391     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 392     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 393     real             *charge;
 394     int              nvdwtype;
 395     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 396     int              *vdwtype;
 397     real             *vdwparam;
 398     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 399     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 400     __m128i          ewitab;
 401     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 402     real             *ewtab;
 403     __m128d          dummy_mask,cutoff_mask;
 404     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 405     __m128d          one     = _mm_set1_pd(1.0);
 406     __m128d          two     = _mm_set1_pd(2.0);
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     facel            = _mm_set1_pd(fr->ic->epsfac);
 419     charge           = mdatoms->chargeA;
 420     nvdwtype         = fr->ntype;
 421     vdwparam         = fr->nbfp;
 422     vdwtype          = mdatoms->typeA;
 423
 424     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 425     ewtab            = fr->ic->tabq_coul_F;
 426     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 427     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 428
 429     /* Avoid stupid compiler warnings */
 430     jnrA = jnrB = 0;
 431     j_coord_offsetA = 0;
 432     j_coord_offsetB = 0;
 433
 434     outeriter        = 0;
 435     inneriter        = 0;
 436
 437     /* Start outer loop over neighborlists */
 438     for(iidx=0; iidx<nri; iidx++)
 439     {
 440         /* Load shift vector for this list */
 441         i_shift_offset   = DIM*shiftidx[iidx];
 442
 443         /* Load limits for loop over neighbors */
 444         j_index_start    = jindex[iidx];
 445         j_index_end      = jindex[iidx+1];
 446
 447         /* Get outer coordinate index */
 448         inr              = iinr[iidx];
 449         i_coord_offset   = DIM*inr;
 450
 451         /* Load i particle coords and add shift vector */
 452         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 453
 454         fix0             = _mm_setzero_pd();
 455         fiy0             = _mm_setzero_pd();
 456         fiz0             = _mm_setzero_pd();
 457
 458         /* Load parameters for i particles */
 459         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 460         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 461
 462         /* Start inner kernel loop */
 463         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrA             = jjnr[jidx];
 468             jnrB             = jjnr[jidx+1];
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471
 472             /* load j atom coordinates */
 473             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _mm_sub_pd(ix0,jx0);
 478             dy00             = _mm_sub_pd(iy0,jy0);
 479             dz00             = _mm_sub_pd(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483
 484             rinv00           = sse41_invsqrt_d(rsq00);
 485
 486             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 490             vdwjidx0A        = 2*vdwtype[jnrA+0];
 491             vdwjidx0B        = 2*vdwtype[jnrB+0];
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r00              = _mm_mul_pd(rsq00,rinv00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq00             = _mm_mul_pd(iq0,jq0);
 501             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 502                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 503
 504             /* EWALD ELECTROSTATICS */
 505
 506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 507             ewrt             = _mm_mul_pd(r00,ewtabscale);
 508             ewitab           = _mm_cvttpd_epi32(ewrt);
 509             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 510             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 511                                          &ewtabF,&ewtabFn);
 512             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 513             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 514
 515             /* LENNARD-JONES DISPERSION/REPULSION */
 516
 517             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 518             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 519
 520             fscal            = _mm_add_pd(felec,fvdw);
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm_mul_pd(fscal,dx00);
 524             ty               = _mm_mul_pd(fscal,dy00);
 525             tz               = _mm_mul_pd(fscal,dz00);
 526
 527             /* Update vectorial force */
 528             fix0             = _mm_add_pd(fix0,tx);
 529             fiy0             = _mm_add_pd(fiy0,ty);
 530             fiz0             = _mm_add_pd(fiz0,tz);
 531
 532             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 533
 534             /* Inner loop uses 43 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             jnrA             = jjnr[jidx];
 541             j_coord_offsetA  = DIM*jnrA;
 542
 543             /* load j atom coordinates */
 544             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_pd(ix0,jx0);
 549             dy00             = _mm_sub_pd(iy0,jy0);
 550             dz00             = _mm_sub_pd(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 554
 555             rinv00           = sse41_invsqrt_d(rsq00);
 556
 557             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = _mm_load_sd(charge+jnrA+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             r00              = _mm_mul_pd(rsq00,rinv00);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq00             = _mm_mul_pd(iq0,jq0);
 571             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 576             ewrt             = _mm_mul_pd(r00,ewtabscale);
 577             ewitab           = _mm_cvttpd_epi32(ewrt);
 578             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 579             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 580             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 581             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 582
 583             /* LENNARD-JONES DISPERSION/REPULSION */
 584
 585             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 586             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 587
 588             fscal            = _mm_add_pd(felec,fvdw);
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_pd(fscal,dx00);
 594             ty               = _mm_mul_pd(fscal,dy00);
 595             tz               = _mm_mul_pd(fscal,dz00);
 596
 597             /* Update vectorial force */
 598             fix0             = _mm_add_pd(fix0,tx);
 599             fiy0             = _mm_add_pd(fiy0,ty);
 600             fiz0             = _mm_add_pd(fiz0,tz);
 601
 602             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 603
 604             /* Inner loop uses 43 flops */
 605         }
 606
 607         /* End of innermost loop */
 608
 609         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 610                                               f+i_coord_offset,fshift+i_shift_offset);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 7 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*43);
 624 }