src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          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     krf              = _mm_set1_pd(fr->ic->k_rf);
 117     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_pd(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_pd();
 152         fiy0             = _mm_setzero_pd();
 153         fiz0             = _mm_setzero_pd();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_pd();
 161         vvdwsum          = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             r00              = _mm_mul_pd(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_pd(iq0,jq0);
 202             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 204
 205             /* Calculate table index by multiplying r with table scale and truncate to integer */
 206             rt               = _mm_mul_pd(r00,vftabscale);
 207             vfitab           = _mm_cvttpd_epi32(rt);
 208             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 209             vfitab           = _mm_slli_epi32(vfitab,3);
 210
 211             /* REACTION-FIELD ELECTROSTATICS */
 212             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 213             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 214
 215             /* CUBIC SPLINE TABLE DISPERSION */
 216             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 217             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 218             GMX_MM_TRANSPOSE2_PD(Y,F);
 219             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 220             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 221             GMX_MM_TRANSPOSE2_PD(G,H);
 222             Heps             = _mm_mul_pd(vfeps,H);
 223             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 224             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 225             vvdw6            = _mm_mul_pd(c6_00,VV);
 226             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 227             fvdw6            = _mm_mul_pd(c6_00,FF);
 228
 229             /* CUBIC SPLINE TABLE REPULSION */
 230             vfitab           = _mm_add_epi32(vfitab,ifour);
 231             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 232             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 233             GMX_MM_TRANSPOSE2_PD(Y,F);
 234             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 235             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 236             GMX_MM_TRANSPOSE2_PD(G,H);
 237             Heps             = _mm_mul_pd(vfeps,H);
 238             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 239             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 240             vvdw12           = _mm_mul_pd(c12_00,VV);
 241             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 242             fvdw12           = _mm_mul_pd(c12_00,FF);
 243             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 244             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 249
 250             fscal            = _mm_add_pd(felec,fvdw);
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_pd(fscal,dx00);
 254             ty               = _mm_mul_pd(fscal,dy00);
 255             tz               = _mm_mul_pd(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_pd(fix0,tx);
 259             fiy0             = _mm_add_pd(fiy0,ty);
 260             fiz0             = _mm_add_pd(fiz0,tz);
 261
 262             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 263
 264             /* Inner loop uses 67 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             jnrA             = jjnr[jidx];
 271             j_coord_offsetA  = DIM*jnrA;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_pd(ix0,jx0);
 279             dy00             = _mm_sub_pd(iy0,jy0);
 280             dz00             = _mm_sub_pd(iz0,jz0);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 284
 285             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 286
 287             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 288
 289             /* Load parameters for j particles */
 290             jq0              = _mm_load_sd(charge+jnrA+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _mm_mul_pd(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _mm_mul_pd(iq0,jq0);
 301             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 302
 303             /* Calculate table index by multiplying r with table scale and truncate to integer */
 304             rt               = _mm_mul_pd(r00,vftabscale);
 305             vfitab           = _mm_cvttpd_epi32(rt);
 306             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 307             vfitab           = _mm_slli_epi32(vfitab,3);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 311             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 312
 313             /* CUBIC SPLINE TABLE DISPERSION */
 314             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 315             F                = _mm_setzero_pd();
 316             GMX_MM_TRANSPOSE2_PD(Y,F);
 317             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 318             H                = _mm_setzero_pd();
 319             GMX_MM_TRANSPOSE2_PD(G,H);
 320             Heps             = _mm_mul_pd(vfeps,H);
 321             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 322             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 323             vvdw6            = _mm_mul_pd(c6_00,VV);
 324             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 325             fvdw6            = _mm_mul_pd(c6_00,FF);
 326
 327             /* CUBIC SPLINE TABLE REPULSION */
 328             vfitab           = _mm_add_epi32(vfitab,ifour);
 329             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 330             F                = _mm_setzero_pd();
 331             GMX_MM_TRANSPOSE2_PD(Y,F);
 332             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 333             H                = _mm_setzero_pd();
 334             GMX_MM_TRANSPOSE2_PD(G,H);
 335             Heps             = _mm_mul_pd(vfeps,H);
 336             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 337             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 338             vvdw12           = _mm_mul_pd(c12_00,VV);
 339             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 340             fvdw12           = _mm_mul_pd(c12_00,FF);
 341             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 342             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 348             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 349
 350             fscal            = _mm_add_pd(felec,fvdw);
 351
 352             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm_mul_pd(fscal,dx00);
 356             ty               = _mm_mul_pd(fscal,dy00);
 357             tz               = _mm_mul_pd(fscal,dz00);
 358
 359             /* Update vectorial force */
 360             fix0             = _mm_add_pd(fix0,tx);
 361             fiy0             = _mm_add_pd(fiy0,ty);
 362             fiz0             = _mm_add_pd(fiz0,tz);
 363
 364             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 365
 366             /* Inner loop uses 67 flops */
 367         }
 368
 369         /* End of innermost loop */
 370
 371         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 372                                               f+i_coord_offset,fshift+i_shift_offset);
 373
 374         ggid                        = gid[iidx];
 375         /* Update potential energies */
 376         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 377         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 378
 379         /* Increment number of inner iterations */
 380         inneriter                  += j_index_end - j_index_start;
 381
 382         /* Outer loop uses 9 flops */
 383     }
 384
 385     /* Increment number of outer iterations */
 386     outeriter        += nri;
 387
 388     /* Update outer/inner flops */
 389
 390     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 391 }
 392 /*
 393  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_double
 394  * Electrostatics interaction: ReactionField
 395  * VdW interaction:            CubicSplineTable
 396  * Geometry:                   Particle-Particle
 397  * Calculate force/pot:        Force
 398  */
 399 void
 400 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_double
 401                     (t_nblist                    * gmx_restrict       nlist,
 402                      rvec                        * gmx_restrict          xx,
 403                      rvec                        * gmx_restrict          ff,
 404                      t_forcerec                  * gmx_restrict          fr,
 405                      t_mdatoms                   * gmx_restrict     mdatoms,
 406                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 407                      t_nrnb                      * gmx_restrict        nrnb)
 408 {
 409     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 410      * just 0 for non-waters.
 411      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 412      * jnr indices corresponding to data put in the four positions in the SIMD register.
 413      */
 414     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 415     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 416     int              jnrA,jnrB;
 417     int              j_coord_offsetA,j_coord_offsetB;
 418     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 419     real             rcutoff_scalar;
 420     real             *shiftvec,*fshift,*x,*f;
 421     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B;
 425     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 428     real             *charge;
 429     int              nvdwtype;
 430     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 431     int              *vdwtype;
 432     real             *vdwparam;
 433     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 434     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 435     __m128i          vfitab;
 436     __m128i          ifour       = _mm_set1_epi32(4);
 437     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 438     real             *vftab;
 439     __m128d          dummy_mask,cutoff_mask;
 440     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 441     __m128d          one     = _mm_set1_pd(1.0);
 442     __m128d          two     = _mm_set1_pd(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm_set1_pd(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456     krf              = _mm_set1_pd(fr->ic->k_rf);
 457     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 458     crf              = _mm_set1_pd(fr->ic->c_rf);
 459     nvdwtype         = fr->ntype;
 460     vdwparam         = fr->nbfp;
 461     vdwtype          = mdatoms->typeA;
 462
 463     vftab            = kernel_data->table_vdw->data;
 464     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479
 480         /* Load limits for loop over neighbors */
 481         j_index_start    = jindex[iidx];
 482         j_index_end      = jindex[iidx+1];
 483
 484         /* Get outer coordinate index */
 485         inr              = iinr[iidx];
 486         i_coord_offset   = DIM*inr;
 487
 488         /* Load i particle coords and add shift vector */
 489         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 490
 491         fix0             = _mm_setzero_pd();
 492         fiy0             = _mm_setzero_pd();
 493         fiz0             = _mm_setzero_pd();
 494
 495         /* Load parameters for i particles */
 496         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 497         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 498
 499         /* Start inner kernel loop */
 500         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrA             = jjnr[jidx];
 505             jnrB             = jjnr[jidx+1];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm_sub_pd(ix0,jx0);
 515             dy00             = _mm_sub_pd(iy0,jy0);
 516             dz00             = _mm_sub_pd(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 522
 523             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm_mul_pd(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _mm_mul_pd(iq0,jq0);
 538             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 540
 541             /* Calculate table index by multiplying r with table scale and truncate to integer */
 542             rt               = _mm_mul_pd(r00,vftabscale);
 543             vfitab           = _mm_cvttpd_epi32(rt);
 544             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 545             vfitab           = _mm_slli_epi32(vfitab,3);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 549
 550             /* CUBIC SPLINE TABLE DISPERSION */
 551             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 553             GMX_MM_TRANSPOSE2_PD(Y,F);
 554             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 555             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 556             GMX_MM_TRANSPOSE2_PD(G,H);
 557             Heps             = _mm_mul_pd(vfeps,H);
 558             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 559             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 560             fvdw6            = _mm_mul_pd(c6_00,FF);
 561
 562             /* CUBIC SPLINE TABLE REPULSION */
 563             vfitab           = _mm_add_epi32(vfitab,ifour);
 564             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 565             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 566             GMX_MM_TRANSPOSE2_PD(Y,F);
 567             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 568             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 569             GMX_MM_TRANSPOSE2_PD(G,H);
 570             Heps             = _mm_mul_pd(vfeps,H);
 571             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 572             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 573             fvdw12           = _mm_mul_pd(c12_00,FF);
 574             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 575
 576             fscal            = _mm_add_pd(felec,fvdw);
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_pd(fscal,dx00);
 580             ty               = _mm_mul_pd(fscal,dy00);
 581             tz               = _mm_mul_pd(fscal,dz00);
 582
 583             /* Update vectorial force */
 584             fix0             = _mm_add_pd(fix0,tx);
 585             fiy0             = _mm_add_pd(fiy0,ty);
 586             fiz0             = _mm_add_pd(fiz0,tz);
 587
 588             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 589
 590             /* Inner loop uses 54 flops */
 591         }
 592
 593         if(jidx<j_index_end)
 594         {
 595
 596             jnrA             = jjnr[jidx];
 597             j_coord_offsetA  = DIM*jnrA;
 598
 599             /* load j atom coordinates */
 600             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 601                                               &jx0,&jy0,&jz0);
 602
 603             /* Calculate displacement vector */
 604             dx00             = _mm_sub_pd(ix0,jx0);
 605             dy00             = _mm_sub_pd(iy0,jy0);
 606             dz00             = _mm_sub_pd(iz0,jz0);
 607
 608             /* Calculate squared distance and things based on it */
 609             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 610
 611             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 612
 613             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 614
 615             /* Load parameters for j particles */
 616             jq0              = _mm_load_sd(charge+jnrA+0);
 617             vdwjidx0A        = 2*vdwtype[jnrA+0];
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r00              = _mm_mul_pd(rsq00,rinv00);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm_mul_pd(iq0,jq0);
 627             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 628
 629             /* Calculate table index by multiplying r with table scale and truncate to integer */
 630             rt               = _mm_mul_pd(r00,vftabscale);
 631             vfitab           = _mm_cvttpd_epi32(rt);
 632             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 633             vfitab           = _mm_slli_epi32(vfitab,3);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 637
 638             /* CUBIC SPLINE TABLE DISPERSION */
 639             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 640             F                = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(Y,F);
 642             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 643             H                = _mm_setzero_pd();
 644             GMX_MM_TRANSPOSE2_PD(G,H);
 645             Heps             = _mm_mul_pd(vfeps,H);
 646             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 647             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 648             fvdw6            = _mm_mul_pd(c6_00,FF);
 649
 650             /* CUBIC SPLINE TABLE REPULSION */
 651             vfitab           = _mm_add_epi32(vfitab,ifour);
 652             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 653             F                = _mm_setzero_pd();
 654             GMX_MM_TRANSPOSE2_PD(Y,F);
 655             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 656             H                = _mm_setzero_pd();
 657             GMX_MM_TRANSPOSE2_PD(G,H);
 658             Heps             = _mm_mul_pd(vfeps,H);
 659             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 660             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 661             fvdw12           = _mm_mul_pd(c12_00,FF);
 662             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 663
 664             fscal            = _mm_add_pd(felec,fvdw);
 665
 666             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_pd(fscal,dx00);
 670             ty               = _mm_mul_pd(fscal,dy00);
 671             tz               = _mm_mul_pd(fscal,dz00);
 672
 673             /* Update vectorial force */
 674             fix0             = _mm_add_pd(fix0,tx);
 675             fiy0             = _mm_add_pd(fiy0,ty);
 676             fiz0             = _mm_add_pd(fiz0,tz);
 677
 678             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 679
 680             /* Inner loop uses 54 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         /* Increment number of inner iterations */
 689         inneriter                  += j_index_end - j_index_start;
 690
 691         /* Outer loop uses 7 flops */
 692     }
 693
 694     /* Increment number of outer iterations */
 695     outeriter        += nri;
 696
 697     /* Update outer/inner flops */
 698
 699     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 700 }