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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_pd();
 148         fiy0             = _mm_setzero_pd();
 149         fiz0             = _mm_setzero_pd();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157         vvdwsum          = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 182
 183             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _mm_mul_pd(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _mm_mul_pd(iq0,jq0);
 198             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* EWALD ELECTROSTATICS */
 202
 203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 204             ewrt             = _mm_mul_pd(r00,ewtabscale);
 205             ewitab           = _mm_cvttpd_epi32(ewrt);
 206             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 207             ewitab           = _mm_slli_epi32(ewitab,2);
 208             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 209             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 210             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 211             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 212             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 213             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 214             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 215             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 216             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 217             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 218
 219             /* LENNARD-JONES DISPERSION/REPULSION */
 220
 221             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 222             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 223             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 224             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 225             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_pd(velecsum,velec);
 229             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 230
 231             fscal            = _mm_add_pd(felec,fvdw);
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 244
 245             /* Inner loop uses 53 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_pd(ix0,jx0);
 260             dy00             = _mm_sub_pd(iy0,jy0);
 261             dz00             = _mm_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 267
 268             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = _mm_load_sd(charge+jnrA+0);
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _mm_mul_pd(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm_mul_pd(iq0,jq0);
 282             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 283
 284             /* EWALD ELECTROSTATICS */
 285
 286             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 287             ewrt             = _mm_mul_pd(r00,ewtabscale);
 288             ewitab           = _mm_cvttpd_epi32(ewrt);
 289             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 290             ewitab           = _mm_slli_epi32(ewitab,2);
 291             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 292             ewtabD           = _mm_setzero_pd();
 293             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 294             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 295             ewtabFn          = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 297             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 298             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 299             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 300             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 301
 302             /* LENNARD-JONES DISPERSION/REPULSION */
 303
 304             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 305             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 306             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 307             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 308             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 312             velecsum         = _mm_add_pd(velecsum,velec);
 313             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 314             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 315
 316             fscal            = _mm_add_pd(felec,fvdw);
 317
 318             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_pd(fscal,dx00);
 322             ty               = _mm_mul_pd(fscal,dy00);
 323             tz               = _mm_mul_pd(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm_add_pd(fix0,tx);
 327             fiy0             = _mm_add_pd(fiy0,ty);
 328             fiz0             = _mm_add_pd(fiz0,tz);
 329
 330             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 331
 332             /* Inner loop uses 53 flops */
 333         }
 334
 335         /* End of innermost loop */
 336
 337         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 338                                               f+i_coord_offset,fshift+i_shift_offset);
 339
 340         ggid                        = gid[iidx];
 341         /* Update potential energies */
 342         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 343         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 344
 345         /* Increment number of inner iterations */
 346         inneriter                  += j_index_end - j_index_start;
 347
 348         /* Outer loop uses 9 flops */
 349     }
 350
 351     /* Increment number of outer iterations */
 352     outeriter        += nri;
 353
 354     /* Update outer/inner flops */
 355
 356     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*53);
 357 }
 358 /*
 359  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_double
 360  * Electrostatics interaction: Ewald
 361  * VdW interaction:            LennardJones
 362  * Geometry:                   Particle-Particle
 363  * Calculate force/pot:        Force
 364  */
 365 void
 366 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_double
 367                     (t_nblist                    * gmx_restrict       nlist,
 368                      rvec                        * gmx_restrict          xx,
 369                      rvec                        * gmx_restrict          ff,
 370                      t_forcerec                  * gmx_restrict          fr,
 371                      t_mdatoms                   * gmx_restrict     mdatoms,
 372                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 373                      t_nrnb                      * gmx_restrict        nrnb)
 374 {
 375     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 376      * just 0 for non-waters.
 377      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 378      * jnr indices corresponding to data put in the four positions in the SIMD register.
 379      */
 380     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 381     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 382     int              jnrA,jnrB;
 383     int              j_coord_offsetA,j_coord_offsetB;
 384     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 385     real             rcutoff_scalar;
 386     real             *shiftvec,*fshift,*x,*f;
 387     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 388     int              vdwioffset0;
 389     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 390     int              vdwjidx0A,vdwjidx0B;
 391     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 392     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 393     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 394     real             *charge;
 395     int              nvdwtype;
 396     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 397     int              *vdwtype;
 398     real             *vdwparam;
 399     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 400     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 401     __m128i          ewitab;
 402     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 403     real             *ewtab;
 404     __m128d          dummy_mask,cutoff_mask;
 405     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 406     __m128d          one     = _mm_set1_pd(1.0);
 407     __m128d          two     = _mm_set1_pd(2.0);
 408     x                = xx[0];
 409     f                = ff[0];
 410
 411     nri              = nlist->nri;
 412     iinr             = nlist->iinr;
 413     jindex           = nlist->jindex;
 414     jjnr             = nlist->jjnr;
 415     shiftidx         = nlist->shift;
 416     gid              = nlist->gid;
 417     shiftvec         = fr->shift_vec[0];
 418     fshift           = fr->fshift[0];
 419     facel            = _mm_set1_pd(fr->epsfac);
 420     charge           = mdatoms->chargeA;
 421     nvdwtype         = fr->ntype;
 422     vdwparam         = fr->nbfp;
 423     vdwtype          = mdatoms->typeA;
 424
 425     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 426     ewtab            = fr->ic->tabq_coul_F;
 427     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 428     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _mm_setzero_pd();
 456         fiy0             = _mm_setzero_pd();
 457         fiz0             = _mm_setzero_pd();
 458
 459         /* Load parameters for i particles */
 460         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 461         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 462
 463         /* Start inner kernel loop */
 464         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrA             = jjnr[jidx];
 469             jnrB             = jjnr[jidx+1];
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472
 473             /* load j atom coordinates */
 474             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               &jx0,&jy0,&jz0);
 476
 477             /* Calculate displacement vector */
 478             dx00             = _mm_sub_pd(ix0,jx0);
 479             dy00             = _mm_sub_pd(iy0,jy0);
 480             dz00             = _mm_sub_pd(iz0,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 484
 485             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 486
 487             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _mm_mul_pd(rsq00,rinv00);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq00             = _mm_mul_pd(iq0,jq0);
 502             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 503                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 504
 505             /* EWALD ELECTROSTATICS */
 506
 507             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 508             ewrt             = _mm_mul_pd(r00,ewtabscale);
 509             ewitab           = _mm_cvttpd_epi32(ewrt);
 510             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 511             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 512                                          &ewtabF,&ewtabFn);
 513             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 514             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 515
 516             /* LENNARD-JONES DISPERSION/REPULSION */
 517
 518             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 519             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 520
 521             fscal            = _mm_add_pd(felec,fvdw);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm_mul_pd(fscal,dx00);
 525             ty               = _mm_mul_pd(fscal,dy00);
 526             tz               = _mm_mul_pd(fscal,dz00);
 527
 528             /* Update vectorial force */
 529             fix0             = _mm_add_pd(fix0,tx);
 530             fiy0             = _mm_add_pd(fiy0,ty);
 531             fiz0             = _mm_add_pd(fiz0,tz);
 532
 533             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 534
 535             /* Inner loop uses 43 flops */
 536         }
 537
 538         if(jidx<j_index_end)
 539         {
 540
 541             jnrA             = jjnr[jidx];
 542             j_coord_offsetA  = DIM*jnrA;
 543
 544             /* load j atom coordinates */
 545             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 546                                               &jx0,&jy0,&jz0);
 547
 548             /* Calculate displacement vector */
 549             dx00             = _mm_sub_pd(ix0,jx0);
 550             dy00             = _mm_sub_pd(iy0,jy0);
 551             dz00             = _mm_sub_pd(iz0,jz0);
 552
 553             /* Calculate squared distance and things based on it */
 554             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 555
 556             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 557
 558             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 559
 560             /* Load parameters for j particles */
 561             jq0              = _mm_load_sd(charge+jnrA+0);
 562             vdwjidx0A        = 2*vdwtype[jnrA+0];
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             r00              = _mm_mul_pd(rsq00,rinv00);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq00             = _mm_mul_pd(iq0,jq0);
 572             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 573
 574             /* EWALD ELECTROSTATICS */
 575
 576             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 577             ewrt             = _mm_mul_pd(r00,ewtabscale);
 578             ewitab           = _mm_cvttpd_epi32(ewrt);
 579             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 580             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 581             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 582             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 583
 584             /* LENNARD-JONES DISPERSION/REPULSION */
 585
 586             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 587             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 588
 589             fscal            = _mm_add_pd(felec,fvdw);
 590
 591             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_pd(fscal,dx00);
 595             ty               = _mm_mul_pd(fscal,dy00);
 596             tz               = _mm_mul_pd(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm_add_pd(fix0,tx);
 600             fiy0             = _mm_add_pd(fiy0,ty);
 601             fiz0             = _mm_add_pd(fiz0,tz);
 602
 603             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 604
 605             /* Inner loop uses 43 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 611                                               f+i_coord_offset,fshift+i_shift_offset);
 612
 613         /* Increment number of inner iterations */
 614         inneriter                  += j_index_end - j_index_start;
 615
 616         /* Outer loop uses 7 flops */
 617     }
 618
 619     /* Increment number of outer iterations */
 620     outeriter        += nri;
 621
 622     /* Update outer/inner flops */
 623
 624     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*43);
 625 }