src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_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 "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_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_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          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_vdw->data;
 119     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->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           = gmx_mm_invsqrt_pd(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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 201             rt               = _mm_mul_pd(r00,vftabscale);
 202             vfitab           = _mm_cvttpd_epi32(rt);
 203             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 204             vfitab           = _mm_slli_epi32(vfitab,3);
 205
 206             /* COULOMB ELECTROSTATICS */
 207             velec            = _mm_mul_pd(qq00,rinv00);
 208             felec            = _mm_mul_pd(velec,rinvsq00);
 209
 210             /* CUBIC SPLINE TABLE DISPERSION */
 211             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 212             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 213             GMX_MM_TRANSPOSE2_PD(Y,F);
 214             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 215             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 216             GMX_MM_TRANSPOSE2_PD(G,H);
 217             Heps             = _mm_mul_pd(vfeps,H);
 218             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 219             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 220             vvdw6            = _mm_mul_pd(c6_00,VV);
 221             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 222             fvdw6            = _mm_mul_pd(c6_00,FF);
 223
 224             /* CUBIC SPLINE TABLE REPULSION */
 225             vfitab           = _mm_add_epi32(vfitab,ifour);
 226             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 227             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(Y,F);
 229             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 230             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(G,H);
 232             Heps             = _mm_mul_pd(vfeps,H);
 233             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 234             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 235             vvdw12           = _mm_mul_pd(c12_00,VV);
 236             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 237             fvdw12           = _mm_mul_pd(c12_00,FF);
 238             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 239             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velecsum         = _mm_add_pd(velecsum,velec);
 243             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 244
 245             fscal            = _mm_add_pd(felec,fvdw);
 246
 247             /* Calculate temporary vectorial force */
 248             tx               = _mm_mul_pd(fscal,dx00);
 249             ty               = _mm_mul_pd(fscal,dy00);
 250             tz               = _mm_mul_pd(fscal,dz00);
 251
 252             /* Update vectorial force */
 253             fix0             = _mm_add_pd(fix0,tx);
 254             fiy0             = _mm_add_pd(fiy0,ty);
 255             fiz0             = _mm_add_pd(fiz0,tz);
 256
 257             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 258
 259             /* Inner loop uses 63 flops */
 260         }
 261
 262         if(jidx<j_index_end)
 263         {
 264
 265             jnrA             = jjnr[jidx];
 266             j_coord_offsetA  = DIM*jnrA;
 267
 268             /* load j atom coordinates */
 269             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 270                                               &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm_sub_pd(ix0,jx0);
 274             dy00             = _mm_sub_pd(iy0,jy0);
 275             dz00             = _mm_sub_pd(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 281
 282             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = _mm_load_sd(charge+jnrA+0);
 286             vdwjidx0A        = 2*vdwtype[jnrA+0];
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r00              = _mm_mul_pd(rsq00,rinv00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _mm_mul_pd(iq0,jq0);
 296             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 297
 298             /* Calculate table index by multiplying r with table scale and truncate to integer */
 299             rt               = _mm_mul_pd(r00,vftabscale);
 300             vfitab           = _mm_cvttpd_epi32(rt);
 301             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 302             vfitab           = _mm_slli_epi32(vfitab,3);
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _mm_mul_pd(qq00,rinv00);
 306             felec            = _mm_mul_pd(velec,rinvsq00);
 307
 308             /* CUBIC SPLINE TABLE DISPERSION */
 309             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 310             F                = _mm_setzero_pd();
 311             GMX_MM_TRANSPOSE2_PD(Y,F);
 312             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 313             H                = _mm_setzero_pd();
 314             GMX_MM_TRANSPOSE2_PD(G,H);
 315             Heps             = _mm_mul_pd(vfeps,H);
 316             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 317             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 318             vvdw6            = _mm_mul_pd(c6_00,VV);
 319             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 320             fvdw6            = _mm_mul_pd(c6_00,FF);
 321
 322             /* CUBIC SPLINE TABLE REPULSION */
 323             vfitab           = _mm_add_epi32(vfitab,ifour);
 324             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 325             F                = _mm_setzero_pd();
 326             GMX_MM_TRANSPOSE2_PD(Y,F);
 327             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 328             H                = _mm_setzero_pd();
 329             GMX_MM_TRANSPOSE2_PD(G,H);
 330             Heps             = _mm_mul_pd(vfeps,H);
 331             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 332             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 333             vvdw12           = _mm_mul_pd(c12_00,VV);
 334             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 335             fvdw12           = _mm_mul_pd(c12_00,FF);
 336             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 337             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 341             velecsum         = _mm_add_pd(velecsum,velec);
 342             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 343             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 344
 345             fscal            = _mm_add_pd(felec,fvdw);
 346
 347             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_pd(fscal,dx00);
 351             ty               = _mm_mul_pd(fscal,dy00);
 352             tz               = _mm_mul_pd(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm_add_pd(fix0,tx);
 356             fiy0             = _mm_add_pd(fiy0,ty);
 357             fiz0             = _mm_add_pd(fiz0,tz);
 358
 359             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 360
 361             /* Inner loop uses 63 flops */
 362         }
 363
 364         /* End of innermost loop */
 365
 366         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 367                                               f+i_coord_offset,fshift+i_shift_offset);
 368
 369         ggid                        = gid[iidx];
 370         /* Update potential energies */
 371         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 372         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 9 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 389  * Electrostatics interaction: Coulomb
 390  * VdW interaction:            CubicSplineTable
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 396                     (t_nblist                    * gmx_restrict       nlist,
 397                      rvec                        * gmx_restrict          xx,
 398                      rvec                        * gmx_restrict          ff,
 399                      t_forcerec                  * gmx_restrict          fr,
 400                      t_mdatoms                   * gmx_restrict     mdatoms,
 401                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 402                      t_nrnb                      * gmx_restrict        nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB;
 412     int              j_coord_offsetA,j_coord_offsetB;
 413     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 414     real             rcutoff_scalar;
 415     real             *shiftvec,*fshift,*x,*f;
 416     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 417     int              vdwioffset0;
 418     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 419     int              vdwjidx0A,vdwjidx0B;
 420     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 421     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 422     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 423     real             *charge;
 424     int              nvdwtype;
 425     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 426     int              *vdwtype;
 427     real             *vdwparam;
 428     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 429     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 430     __m128i          vfitab;
 431     __m128i          ifour       = _mm_set1_epi32(4);
 432     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 433     real             *vftab;
 434     __m128d          dummy_mask,cutoff_mask;
 435     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 436     __m128d          one     = _mm_set1_pd(1.0);
 437     __m128d          two     = _mm_set1_pd(2.0);
 438     x                = xx[0];
 439     f                = ff[0];
 440
 441     nri              = nlist->nri;
 442     iinr             = nlist->iinr;
 443     jindex           = nlist->jindex;
 444     jjnr             = nlist->jjnr;
 445     shiftidx         = nlist->shift;
 446     gid              = nlist->gid;
 447     shiftvec         = fr->shift_vec[0];
 448     fshift           = fr->fshift[0];
 449     facel            = _mm_set1_pd(fr->epsfac);
 450     charge           = mdatoms->chargeA;
 451     nvdwtype         = fr->ntype;
 452     vdwparam         = fr->nbfp;
 453     vdwtype          = mdatoms->typeA;
 454
 455     vftab            = kernel_data->table_vdw->data;
 456     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 457
 458     /* Avoid stupid compiler warnings */
 459     jnrA = jnrB = 0;
 460     j_coord_offsetA = 0;
 461     j_coord_offsetB = 0;
 462
 463     outeriter        = 0;
 464     inneriter        = 0;
 465
 466     /* Start outer loop over neighborlists */
 467     for(iidx=0; iidx<nri; iidx++)
 468     {
 469         /* Load shift vector for this list */
 470         i_shift_offset   = DIM*shiftidx[iidx];
 471
 472         /* Load limits for loop over neighbors */
 473         j_index_start    = jindex[iidx];
 474         j_index_end      = jindex[iidx+1];
 475
 476         /* Get outer coordinate index */
 477         inr              = iinr[iidx];
 478         i_coord_offset   = DIM*inr;
 479
 480         /* Load i particle coords and add shift vector */
 481         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 482
 483         fix0             = _mm_setzero_pd();
 484         fiy0             = _mm_setzero_pd();
 485         fiz0             = _mm_setzero_pd();
 486
 487         /* Load parameters for i particles */
 488         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 489         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 490
 491         /* Start inner kernel loop */
 492         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrA             = jjnr[jidx];
 497             jnrB             = jjnr[jidx+1];
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500
 501             /* load j atom coordinates */
 502             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _mm_sub_pd(ix0,jx0);
 507             dy00             = _mm_sub_pd(iy0,jy0);
 508             dz00             = _mm_sub_pd(iz0,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 512
 513             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 514
 515             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520             vdwjidx0B        = 2*vdwtype[jnrB+0];
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r00              = _mm_mul_pd(rsq00,rinv00);
 527
 528             /* Compute parameters for interactions between i and j atoms */
 529             qq00             = _mm_mul_pd(iq0,jq0);
 530             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 531                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm_mul_pd(r00,vftabscale);
 535             vfitab           = _mm_cvttpd_epi32(rt);
 536             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 537             vfitab           = _mm_slli_epi32(vfitab,3);
 538
 539             /* COULOMB ELECTROSTATICS */
 540             velec            = _mm_mul_pd(qq00,rinv00);
 541             felec            = _mm_mul_pd(velec,rinvsq00);
 542
 543             /* CUBIC SPLINE TABLE DISPERSION */
 544             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 545             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 546             GMX_MM_TRANSPOSE2_PD(Y,F);
 547             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 548             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 549             GMX_MM_TRANSPOSE2_PD(G,H);
 550             Heps             = _mm_mul_pd(vfeps,H);
 551             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 552             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 553             fvdw6            = _mm_mul_pd(c6_00,FF);
 554
 555             /* CUBIC SPLINE TABLE REPULSION */
 556             vfitab           = _mm_add_epi32(vfitab,ifour);
 557             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 558             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 559             GMX_MM_TRANSPOSE2_PD(Y,F);
 560             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 561             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 562             GMX_MM_TRANSPOSE2_PD(G,H);
 563             Heps             = _mm_mul_pd(vfeps,H);
 564             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 565             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 566             fvdw12           = _mm_mul_pd(c12_00,FF);
 567             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 568
 569             fscal            = _mm_add_pd(felec,fvdw);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm_mul_pd(fscal,dx00);
 573             ty               = _mm_mul_pd(fscal,dy00);
 574             tz               = _mm_mul_pd(fscal,dz00);
 575
 576             /* Update vectorial force */
 577             fix0             = _mm_add_pd(fix0,tx);
 578             fiy0             = _mm_add_pd(fiy0,ty);
 579             fiz0             = _mm_add_pd(fiz0,tz);
 580
 581             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 582
 583             /* Inner loop uses 54 flops */
 584         }
 585
 586         if(jidx<j_index_end)
 587         {
 588
 589             jnrA             = jjnr[jidx];
 590             j_coord_offsetA  = DIM*jnrA;
 591
 592             /* load j atom coordinates */
 593             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 594                                               &jx0,&jy0,&jz0);
 595
 596             /* Calculate displacement vector */
 597             dx00             = _mm_sub_pd(ix0,jx0);
 598             dy00             = _mm_sub_pd(iy0,jy0);
 599             dz00             = _mm_sub_pd(iz0,jz0);
 600
 601             /* Calculate squared distance and things based on it */
 602             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 603
 604             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 605
 606             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 607
 608             /* Load parameters for j particles */
 609             jq0              = _mm_load_sd(charge+jnrA+0);
 610             vdwjidx0A        = 2*vdwtype[jnrA+0];
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r00              = _mm_mul_pd(rsq00,rinv00);
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq00             = _mm_mul_pd(iq0,jq0);
 620             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 621
 622             /* Calculate table index by multiplying r with table scale and truncate to integer */
 623             rt               = _mm_mul_pd(r00,vftabscale);
 624             vfitab           = _mm_cvttpd_epi32(rt);
 625             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 626             vfitab           = _mm_slli_epi32(vfitab,3);
 627
 628             /* COULOMB ELECTROSTATICS */
 629             velec            = _mm_mul_pd(qq00,rinv00);
 630             felec            = _mm_mul_pd(velec,rinvsq00);
 631
 632             /* CUBIC SPLINE TABLE DISPERSION */
 633             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 634             F                = _mm_setzero_pd();
 635             GMX_MM_TRANSPOSE2_PD(Y,F);
 636             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 637             H                = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(G,H);
 639             Heps             = _mm_mul_pd(vfeps,H);
 640             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 641             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 642             fvdw6            = _mm_mul_pd(c6_00,FF);
 643
 644             /* CUBIC SPLINE TABLE REPULSION */
 645             vfitab           = _mm_add_epi32(vfitab,ifour);
 646             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 647             F                = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(Y,F);
 649             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 650             H                = _mm_setzero_pd();
 651             GMX_MM_TRANSPOSE2_PD(G,H);
 652             Heps             = _mm_mul_pd(vfeps,H);
 653             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 654             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 655             fvdw12           = _mm_mul_pd(c12_00,FF);
 656             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 657
 658             fscal            = _mm_add_pd(felec,fvdw);
 659
 660             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_pd(fscal,dx00);
 664             ty               = _mm_mul_pd(fscal,dy00);
 665             tz               = _mm_mul_pd(fscal,dz00);
 666
 667             /* Update vectorial force */
 668             fix0             = _mm_add_pd(fix0,tx);
 669             fiy0             = _mm_add_pd(fiy0,ty);
 670             fiz0             = _mm_add_pd(fiz0,tz);
 671
 672             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 673
 674             /* Inner loop uses 54 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 680                                               f+i_coord_offset,fshift+i_shift_offset);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 7 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 694 }