src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "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_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse4_1_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          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_elec->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_pd();
 149         fiy0             = _mm_setzero_pd();
 150         fiz0             = _mm_setzero_pd();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 183
 184             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             r00              = _mm_mul_pd(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm_mul_pd(iq0,jq0);
 199             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 200                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 201
 202             /* Calculate table index by multiplying r with table scale and truncate to integer */
 203             rt               = _mm_mul_pd(r00,vftabscale);
 204             vfitab           = _mm_cvttpd_epi32(rt);
 205             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 206             vfitab           = _mm_slli_epi32(vfitab,2);
 207
 208             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 209             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 210             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 211             GMX_MM_TRANSPOSE2_PD(Y,F);
 212             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 213             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 214             GMX_MM_TRANSPOSE2_PD(G,H);
 215             Heps             = _mm_mul_pd(vfeps,H);
 216             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 217             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 218             velec            = _mm_mul_pd(qq00,VV);
 219             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 220             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 221
 222             /* LENNARD-JONES DISPERSION/REPULSION */
 223
 224             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 225             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 226             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 227             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 228             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velecsum         = _mm_add_pd(velecsum,velec);
 232             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 233
 234             fscal            = _mm_add_pd(felec,fvdw);
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_pd(fscal,dx00);
 238             ty               = _mm_mul_pd(fscal,dy00);
 239             tz               = _mm_mul_pd(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_pd(fix0,tx);
 243             fiy0             = _mm_add_pd(fiy0,ty);
 244             fiz0             = _mm_add_pd(fiz0,tz);
 245
 246             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 247
 248             /* Inner loop uses 56 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             jnrA             = jjnr[jidx];
 255             j_coord_offsetA  = DIM*jnrA;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 259                                               &jx0,&jy0,&jz0);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm_sub_pd(ix0,jx0);
 263             dy00             = _mm_sub_pd(iy0,jy0);
 264             dz00             = _mm_sub_pd(iz0,jz0);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 268
 269             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 270
 271             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 272
 273             /* Load parameters for j particles */
 274             jq0              = _mm_load_sd(charge+jnrA+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             r00              = _mm_mul_pd(rsq00,rinv00);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq00             = _mm_mul_pd(iq0,jq0);
 285             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 286
 287             /* Calculate table index by multiplying r with table scale and truncate to integer */
 288             rt               = _mm_mul_pd(r00,vftabscale);
 289             vfitab           = _mm_cvttpd_epi32(rt);
 290             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 291             vfitab           = _mm_slli_epi32(vfitab,2);
 292
 293             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 294             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 295             F                = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(Y,F);
 297             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 298             H                = _mm_setzero_pd();
 299             GMX_MM_TRANSPOSE2_PD(G,H);
 300             Heps             = _mm_mul_pd(vfeps,H);
 301             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 302             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 303             velec            = _mm_mul_pd(qq00,VV);
 304             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 305             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 306
 307             /* LENNARD-JONES DISPERSION/REPULSION */
 308
 309             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 310             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 311             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 312             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 313             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 317             velecsum         = _mm_add_pd(velecsum,velec);
 318             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 319             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 320
 321             fscal            = _mm_add_pd(felec,fvdw);
 322
 323             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_pd(fscal,dx00);
 327             ty               = _mm_mul_pd(fscal,dy00);
 328             tz               = _mm_mul_pd(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_add_pd(fix0,tx);
 332             fiy0             = _mm_add_pd(fiy0,ty);
 333             fiz0             = _mm_add_pd(fiz0,tz);
 334
 335             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 336
 337             /* Inner loop uses 56 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 343                                               f+i_coord_offset,fshift+i_shift_offset);
 344
 345         ggid                        = gid[iidx];
 346         /* Update potential energies */
 347         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 348         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 9 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse4_1_double
 365  * Electrostatics interaction: CubicSplineTable
 366  * VdW interaction:            LennardJones
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse4_1_double
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      t_forcerec                  * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB;
 388     int              j_coord_offsetA,j_coord_offsetB;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     int              vdwioffset0;
 394     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B;
 396     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 399     real             *charge;
 400     int              nvdwtype;
 401     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 402     int              *vdwtype;
 403     real             *vdwparam;
 404     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 405     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 406     __m128i          vfitab;
 407     __m128i          ifour       = _mm_set1_epi32(4);
 408     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 409     real             *vftab;
 410     __m128d          dummy_mask,cutoff_mask;
 411     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 412     __m128d          one     = _mm_set1_pd(1.0);
 413     __m128d          two     = _mm_set1_pd(2.0);
 414     x                = xx[0];
 415     f                = ff[0];
 416
 417     nri              = nlist->nri;
 418     iinr             = nlist->iinr;
 419     jindex           = nlist->jindex;
 420     jjnr             = nlist->jjnr;
 421     shiftidx         = nlist->shift;
 422     gid              = nlist->gid;
 423     shiftvec         = fr->shift_vec[0];
 424     fshift           = fr->fshift[0];
 425     facel            = _mm_set1_pd(fr->epsfac);
 426     charge           = mdatoms->chargeA;
 427     nvdwtype         = fr->ntype;
 428     vdwparam         = fr->nbfp;
 429     vdwtype          = mdatoms->typeA;
 430
 431     vftab            = kernel_data->table_elec->data;
 432     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 433
 434     /* Avoid stupid compiler warnings */
 435     jnrA = jnrB = 0;
 436     j_coord_offsetA = 0;
 437     j_coord_offsetB = 0;
 438
 439     outeriter        = 0;
 440     inneriter        = 0;
 441
 442     /* Start outer loop over neighborlists */
 443     for(iidx=0; iidx<nri; iidx++)
 444     {
 445         /* Load shift vector for this list */
 446         i_shift_offset   = DIM*shiftidx[iidx];
 447
 448         /* Load limits for loop over neighbors */
 449         j_index_start    = jindex[iidx];
 450         j_index_end      = jindex[iidx+1];
 451
 452         /* Get outer coordinate index */
 453         inr              = iinr[iidx];
 454         i_coord_offset   = DIM*inr;
 455
 456         /* Load i particle coords and add shift vector */
 457         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 458
 459         fix0             = _mm_setzero_pd();
 460         fiy0             = _mm_setzero_pd();
 461         fiz0             = _mm_setzero_pd();
 462
 463         /* Load parameters for i particles */
 464         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 465         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             j_coord_offsetA  = DIM*jnrA;
 475             j_coord_offsetB  = DIM*jnrB;
 476
 477             /* load j atom coordinates */
 478             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 479                                               &jx0,&jy0,&jz0);
 480
 481             /* Calculate displacement vector */
 482             dx00             = _mm_sub_pd(ix0,jx0);
 483             dy00             = _mm_sub_pd(iy0,jy0);
 484             dz00             = _mm_sub_pd(iz0,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 488
 489             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 490
 491             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 492
 493             /* Load parameters for j particles */
 494             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496             vdwjidx0B        = 2*vdwtype[jnrB+0];
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r00              = _mm_mul_pd(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq00             = _mm_mul_pd(iq0,jq0);
 506             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm_mul_pd(r00,vftabscale);
 511             vfitab           = _mm_cvttpd_epi32(rt);
 512             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 513             vfitab           = _mm_slli_epi32(vfitab,2);
 514
 515             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 516             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 517             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 518             GMX_MM_TRANSPOSE2_PD(Y,F);
 519             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 520             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 521             GMX_MM_TRANSPOSE2_PD(G,H);
 522             Heps             = _mm_mul_pd(vfeps,H);
 523             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 524             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 525             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 530             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 531
 532             fscal            = _mm_add_pd(felec,fvdw);
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm_mul_pd(fscal,dx00);
 536             ty               = _mm_mul_pd(fscal,dy00);
 537             tz               = _mm_mul_pd(fscal,dz00);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_add_pd(fix0,tx);
 541             fiy0             = _mm_add_pd(fiy0,ty);
 542             fiz0             = _mm_add_pd(fiz0,tz);
 543
 544             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 545
 546             /* Inner loop uses 47 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             jnrA             = jjnr[jidx];
 553             j_coord_offsetA  = DIM*jnrA;
 554
 555             /* load j atom coordinates */
 556             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm_sub_pd(ix0,jx0);
 561             dy00             = _mm_sub_pd(iy0,jy0);
 562             dz00             = _mm_sub_pd(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 566
 567             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 568
 569             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 570
 571             /* Load parameters for j particles */
 572             jq0              = _mm_load_sd(charge+jnrA+0);
 573             vdwjidx0A        = 2*vdwtype[jnrA+0];
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r00              = _mm_mul_pd(rsq00,rinv00);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq00             = _mm_mul_pd(iq0,jq0);
 583             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 584
 585             /* Calculate table index by multiplying r with table scale and truncate to integer */
 586             rt               = _mm_mul_pd(r00,vftabscale);
 587             vfitab           = _mm_cvttpd_epi32(rt);
 588             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 589             vfitab           = _mm_slli_epi32(vfitab,2);
 590
 591             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 592             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 593             F                = _mm_setzero_pd();
 594             GMX_MM_TRANSPOSE2_PD(Y,F);
 595             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 596             H                = _mm_setzero_pd();
 597             GMX_MM_TRANSPOSE2_PD(G,H);
 598             Heps             = _mm_mul_pd(vfeps,H);
 599             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 600             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 601             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 602
 603             /* LENNARD-JONES DISPERSION/REPULSION */
 604
 605             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 606             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 607
 608             fscal            = _mm_add_pd(felec,fvdw);
 609
 610             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm_mul_pd(fscal,dx00);
 614             ty               = _mm_mul_pd(fscal,dy00);
 615             tz               = _mm_mul_pd(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm_add_pd(fix0,tx);
 619             fiy0             = _mm_add_pd(fiy0,ty);
 620             fiz0             = _mm_add_pd(fiz0,tz);
 621
 622             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 623
 624             /* Inner loop uses 47 flops */
 625         }
 626
 627         /* End of innermost loop */
 628
 629         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 630                                               f+i_coord_offset,fshift+i_shift_offset);
 631
 632         /* Increment number of inner iterations */
 633         inneriter                  += j_index_end - j_index_start;
 634
 635         /* Outer loop uses 7 flops */
 636     }
 637
 638     /* Increment number of outer iterations */
 639     outeriter        += nri;
 640
 641     /* Update outer/inner flops */
 642
 643     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*47);
 644 }