src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec_vdw->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_pd();
 155         vvdwsum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = sse41_invsqrt_d(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             r00              = _mm_mul_pd(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _mm_mul_pd(iq0,jq0);
 194             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 196
 197             /* Calculate table index by multiplying r with table scale and truncate to integer */
 198             rt               = _mm_mul_pd(r00,vftabscale);
 199             vfitab           = _mm_cvttpd_epi32(rt);
 200             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 201             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 202
 203             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 204             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 205             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 206             GMX_MM_TRANSPOSE2_PD(Y,F);
 207             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 208             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 209             GMX_MM_TRANSPOSE2_PD(G,H);
 210             Heps             = _mm_mul_pd(vfeps,H);
 211             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 212             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 213             velec            = _mm_mul_pd(qq00,VV);
 214             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 215             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 216
 217             /* CUBIC SPLINE TABLE DISPERSION */
 218             vfitab           = _mm_add_epi32(vfitab,ifour);
 219             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(Y,F);
 222             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 223             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(G,H);
 225             Heps             = _mm_mul_pd(vfeps,H);
 226             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 227             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 228             vvdw6            = _mm_mul_pd(c6_00,VV);
 229             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 230             fvdw6            = _mm_mul_pd(c6_00,FF);
 231
 232             /* CUBIC SPLINE TABLE REPULSION */
 233             vfitab           = _mm_add_epi32(vfitab,ifour);
 234             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 235             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 236             GMX_MM_TRANSPOSE2_PD(Y,F);
 237             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 238             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 239             GMX_MM_TRANSPOSE2_PD(G,H);
 240             Heps             = _mm_mul_pd(vfeps,H);
 241             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 242             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 243             vvdw12           = _mm_mul_pd(c12_00,VV);
 244             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 245             fvdw12           = _mm_mul_pd(c12_00,FF);
 246             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 247             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm_add_pd(velecsum,velec);
 251             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 252
 253             fscal            = _mm_add_pd(felec,fvdw);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_pd(fscal,dx00);
 257             ty               = _mm_mul_pd(fscal,dy00);
 258             tz               = _mm_mul_pd(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_pd(fix0,tx);
 262             fiy0             = _mm_add_pd(fiy0,ty);
 263             fiz0             = _mm_add_pd(fiz0,tz);
 264
 265             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 266
 267             /* Inner loop uses 73 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = sse41_invsqrt_d(rsq00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             r00              = _mm_mul_pd(rsq00,rinv00);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm_mul_pd(iq0,jq0);
 302             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 303
 304             /* Calculate table index by multiplying r with table scale and truncate to integer */
 305             rt               = _mm_mul_pd(r00,vftabscale);
 306             vfitab           = _mm_cvttpd_epi32(rt);
 307             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 308             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 309
 310             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 311             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 312             F                = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(Y,F);
 314             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 315             H                = _mm_setzero_pd();
 316             GMX_MM_TRANSPOSE2_PD(G,H);
 317             Heps             = _mm_mul_pd(vfeps,H);
 318             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 319             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 320             velec            = _mm_mul_pd(qq00,VV);
 321             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 322             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 323
 324             /* CUBIC SPLINE TABLE DISPERSION */
 325             vfitab           = _mm_add_epi32(vfitab,ifour);
 326             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 327             F                = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(Y,F);
 329             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 330             H                = _mm_setzero_pd();
 331             GMX_MM_TRANSPOSE2_PD(G,H);
 332             Heps             = _mm_mul_pd(vfeps,H);
 333             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 334             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 335             vvdw6            = _mm_mul_pd(c6_00,VV);
 336             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 337             fvdw6            = _mm_mul_pd(c6_00,FF);
 338
 339             /* CUBIC SPLINE TABLE REPULSION */
 340             vfitab           = _mm_add_epi32(vfitab,ifour);
 341             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 342             F                = _mm_setzero_pd();
 343             GMX_MM_TRANSPOSE2_PD(Y,F);
 344             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 345             H                = _mm_setzero_pd();
 346             GMX_MM_TRANSPOSE2_PD(G,H);
 347             Heps             = _mm_mul_pd(vfeps,H);
 348             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 349             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 350             vvdw12           = _mm_mul_pd(c12_00,VV);
 351             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 352             fvdw12           = _mm_mul_pd(c12_00,FF);
 353             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 354             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 358             velecsum         = _mm_add_pd(velecsum,velec);
 359             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 360             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 361
 362             fscal            = _mm_add_pd(felec,fvdw);
 363
 364             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 365
 366             /* Calculate temporary vectorial force */
 367             tx               = _mm_mul_pd(fscal,dx00);
 368             ty               = _mm_mul_pd(fscal,dy00);
 369             tz               = _mm_mul_pd(fscal,dz00);
 370
 371             /* Update vectorial force */
 372             fix0             = _mm_add_pd(fix0,tx);
 373             fiy0             = _mm_add_pd(fiy0,ty);
 374             fiz0             = _mm_add_pd(fiz0,tz);
 375
 376             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 377
 378             /* Inner loop uses 73 flops */
 379         }
 380
 381         /* End of innermost loop */
 382
 383         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 384                                               f+i_coord_offset,fshift+i_shift_offset);
 385
 386         ggid                        = gid[iidx];
 387         /* Update potential energies */
 388         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 389         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 390
 391         /* Increment number of inner iterations */
 392         inneriter                  += j_index_end - j_index_start;
 393
 394         /* Outer loop uses 9 flops */
 395     }
 396
 397     /* Increment number of outer iterations */
 398     outeriter        += nri;
 399
 400     /* Update outer/inner flops */
 401
 402     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 403 }
 404 /*
 405  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse4_1_double
 406  * Electrostatics interaction: CubicSplineTable
 407  * VdW interaction:            CubicSplineTable
 408  * Geometry:                   Particle-Particle
 409  * Calculate force/pot:        Force
 410  */
 411 void
 412 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse4_1_double
 413                     (t_nblist                    * gmx_restrict       nlist,
 414                      rvec                        * gmx_restrict          xx,
 415                      rvec                        * gmx_restrict          ff,
 416                      struct t_forcerec           * gmx_restrict          fr,
 417                      t_mdatoms                   * gmx_restrict     mdatoms,
 418                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 419                      t_nrnb                      * gmx_restrict        nrnb)
 420 {
 421     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 422      * just 0 for non-waters.
 423      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 424      * jnr indices corresponding to data put in the four positions in the SIMD register.
 425      */
 426     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 427     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 428     int              jnrA,jnrB;
 429     int              j_coord_offsetA,j_coord_offsetB;
 430     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 431     real             rcutoff_scalar;
 432     real             *shiftvec,*fshift,*x,*f;
 433     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 434     int              vdwioffset0;
 435     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B;
 437     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     int              nvdwtype;
 442     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 443     int              *vdwtype;
 444     real             *vdwparam;
 445     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 446     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 447     __m128i          vfitab;
 448     __m128i          ifour       = _mm_set1_epi32(4);
 449     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 450     real             *vftab;
 451     __m128d          dummy_mask,cutoff_mask;
 452     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 453     __m128d          one     = _mm_set1_pd(1.0);
 454     __m128d          two     = _mm_set1_pd(2.0);
 455     x                = xx[0];
 456     f                = ff[0];
 457
 458     nri              = nlist->nri;
 459     iinr             = nlist->iinr;
 460     jindex           = nlist->jindex;
 461     jjnr             = nlist->jjnr;
 462     shiftidx         = nlist->shift;
 463     gid              = nlist->gid;
 464     shiftvec         = fr->shift_vec[0];
 465     fshift           = fr->fshift[0];
 466     facel            = _mm_set1_pd(fr->ic->epsfac);
 467     charge           = mdatoms->chargeA;
 468     nvdwtype         = fr->ntype;
 469     vdwparam         = fr->nbfp;
 470     vdwtype          = mdatoms->typeA;
 471
 472     vftab            = kernel_data->table_elec_vdw->data;
 473     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 474
 475     /* Avoid stupid compiler warnings */
 476     jnrA = jnrB = 0;
 477     j_coord_offsetA = 0;
 478     j_coord_offsetB = 0;
 479
 480     outeriter        = 0;
 481     inneriter        = 0;
 482
 483     /* Start outer loop over neighborlists */
 484     for(iidx=0; iidx<nri; iidx++)
 485     {
 486         /* Load shift vector for this list */
 487         i_shift_offset   = DIM*shiftidx[iidx];
 488
 489         /* Load limits for loop over neighbors */
 490         j_index_start    = jindex[iidx];
 491         j_index_end      = jindex[iidx+1];
 492
 493         /* Get outer coordinate index */
 494         inr              = iinr[iidx];
 495         i_coord_offset   = DIM*inr;
 496
 497         /* Load i particle coords and add shift vector */
 498         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 499
 500         fix0             = _mm_setzero_pd();
 501         fiy0             = _mm_setzero_pd();
 502         fiz0             = _mm_setzero_pd();
 503
 504         /* Load parameters for i particles */
 505         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 506         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 507
 508         /* Start inner kernel loop */
 509         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 510         {
 511
 512             /* Get j neighbor index, and coordinate index */
 513             jnrA             = jjnr[jidx];
 514             jnrB             = jjnr[jidx+1];
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517
 518             /* load j atom coordinates */
 519             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 520                                               &jx0,&jy0,&jz0);
 521
 522             /* Calculate displacement vector */
 523             dx00             = _mm_sub_pd(ix0,jx0);
 524             dy00             = _mm_sub_pd(iy0,jy0);
 525             dz00             = _mm_sub_pd(iz0,jz0);
 526
 527             /* Calculate squared distance and things based on it */
 528             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 529
 530             rinv00           = sse41_invsqrt_d(rsq00);
 531
 532             /* Load parameters for j particles */
 533             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 534             vdwjidx0A        = 2*vdwtype[jnrA+0];
 535             vdwjidx0B        = 2*vdwtype[jnrB+0];
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r00              = _mm_mul_pd(rsq00,rinv00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq00             = _mm_mul_pd(iq0,jq0);
 545             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 546                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 547
 548             /* Calculate table index by multiplying r with table scale and truncate to integer */
 549             rt               = _mm_mul_pd(r00,vftabscale);
 550             vfitab           = _mm_cvttpd_epi32(rt);
 551             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 552             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 553
 554             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 555             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 556             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 557             GMX_MM_TRANSPOSE2_PD(Y,F);
 558             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 559             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 560             GMX_MM_TRANSPOSE2_PD(G,H);
 561             Heps             = _mm_mul_pd(vfeps,H);
 562             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 563             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 564             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 565
 566             /* CUBIC SPLINE TABLE DISPERSION */
 567             vfitab           = _mm_add_epi32(vfitab,ifour);
 568             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 569             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 570             GMX_MM_TRANSPOSE2_PD(Y,F);
 571             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 572             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 573             GMX_MM_TRANSPOSE2_PD(G,H);
 574             Heps             = _mm_mul_pd(vfeps,H);
 575             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 576             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 577             fvdw6            = _mm_mul_pd(c6_00,FF);
 578
 579             /* CUBIC SPLINE TABLE REPULSION */
 580             vfitab           = _mm_add_epi32(vfitab,ifour);
 581             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 582             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 583             GMX_MM_TRANSPOSE2_PD(Y,F);
 584             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 585             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 586             GMX_MM_TRANSPOSE2_PD(G,H);
 587             Heps             = _mm_mul_pd(vfeps,H);
 588             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 589             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 590             fvdw12           = _mm_mul_pd(c12_00,FF);
 591             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 592
 593             fscal            = _mm_add_pd(felec,fvdw);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx00);
 597             ty               = _mm_mul_pd(fscal,dy00);
 598             tz               = _mm_mul_pd(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_pd(fix0,tx);
 602             fiy0             = _mm_add_pd(fiy0,ty);
 603             fiz0             = _mm_add_pd(fiz0,tz);
 604
 605             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 606
 607             /* Inner loop uses 61 flops */
 608         }
 609
 610         if(jidx<j_index_end)
 611         {
 612
 613             jnrA             = jjnr[jidx];
 614             j_coord_offsetA  = DIM*jnrA;
 615
 616             /* load j atom coordinates */
 617             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 618                                               &jx0,&jy0,&jz0);
 619
 620             /* Calculate displacement vector */
 621             dx00             = _mm_sub_pd(ix0,jx0);
 622             dy00             = _mm_sub_pd(iy0,jy0);
 623             dz00             = _mm_sub_pd(iz0,jz0);
 624
 625             /* Calculate squared distance and things based on it */
 626             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 627
 628             rinv00           = sse41_invsqrt_d(rsq00);
 629
 630             /* Load parameters for j particles */
 631             jq0              = _mm_load_sd(charge+jnrA+0);
 632             vdwjidx0A        = 2*vdwtype[jnrA+0];
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             r00              = _mm_mul_pd(rsq00,rinv00);
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq00             = _mm_mul_pd(iq0,jq0);
 642             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 643
 644             /* Calculate table index by multiplying r with table scale and truncate to integer */
 645             rt               = _mm_mul_pd(r00,vftabscale);
 646             vfitab           = _mm_cvttpd_epi32(rt);
 647             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 648             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 649
 650             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 651             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 652             F                = _mm_setzero_pd();
 653             GMX_MM_TRANSPOSE2_PD(Y,F);
 654             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 655             H                = _mm_setzero_pd();
 656             GMX_MM_TRANSPOSE2_PD(G,H);
 657             Heps             = _mm_mul_pd(vfeps,H);
 658             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 659             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 660             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 661
 662             /* CUBIC SPLINE TABLE DISPERSION */
 663             vfitab           = _mm_add_epi32(vfitab,ifour);
 664             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 665             F                = _mm_setzero_pd();
 666             GMX_MM_TRANSPOSE2_PD(Y,F);
 667             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 668             H                = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(G,H);
 670             Heps             = _mm_mul_pd(vfeps,H);
 671             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 672             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 673             fvdw6            = _mm_mul_pd(c6_00,FF);
 674
 675             /* CUBIC SPLINE TABLE REPULSION */
 676             vfitab           = _mm_add_epi32(vfitab,ifour);
 677             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 678             F                = _mm_setzero_pd();
 679             GMX_MM_TRANSPOSE2_PD(Y,F);
 680             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 681             H                = _mm_setzero_pd();
 682             GMX_MM_TRANSPOSE2_PD(G,H);
 683             Heps             = _mm_mul_pd(vfeps,H);
 684             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 685             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 686             fvdw12           = _mm_mul_pd(c12_00,FF);
 687             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 688
 689             fscal            = _mm_add_pd(felec,fvdw);
 690
 691             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_pd(fscal,dx00);
 695             ty               = _mm_mul_pd(fscal,dy00);
 696             tz               = _mm_mul_pd(fscal,dz00);
 697
 698             /* Update vectorial force */
 699             fix0             = _mm_add_pd(fix0,tx);
 700             fiy0             = _mm_add_pd(fiy0,ty);
 701             fiz0             = _mm_add_pd(fiz0,tz);
 702
 703             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 704
 705             /* Inner loop uses 61 flops */
 706         }
 707
 708         /* End of innermost loop */
 709
 710         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 711                                               f+i_coord_offset,fshift+i_shift_offset);
 712
 713         /* Increment number of inner iterations */
 714         inneriter                  += j_index_end - j_index_start;
 715
 716         /* Outer loop uses 7 flops */
 717     }
 718
 719     /* Increment number of outer iterations */
 720     outeriter        += nri;
 721
 722     /* Update outer/inner flops */
 723
 724     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 725 }