src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_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     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_elec_vdw->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_pd();
 149         fiy0             = _mm_setzero_pd();
 150         fiz0             = _mm_setzero_pd();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 183
 184             /* 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_cvtepi32_pd(vfitab));
 204             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 205
 206             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 207             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 208             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 209             GMX_MM_TRANSPOSE2_PD(Y,F);
 210             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 211             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 212             GMX_MM_TRANSPOSE2_PD(G,H);
 213             Heps             = _mm_mul_pd(vfeps,H);
 214             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 215             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 216             velec            = _mm_mul_pd(qq00,VV);
 217             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 218             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 219
 220             /* CUBIC SPLINE TABLE DISPERSION */
 221             vfitab           = _mm_add_epi32(vfitab,ifour);
 222             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 223             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 224             GMX_MM_TRANSPOSE2_PD(Y,F);
 225             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 226             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 227             GMX_MM_TRANSPOSE2_PD(G,H);
 228             Heps             = _mm_mul_pd(vfeps,H);
 229             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 230             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 231             vvdw6            = _mm_mul_pd(c6_00,VV);
 232             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 233             fvdw6            = _mm_mul_pd(c6_00,FF);
 234
 235             /* CUBIC SPLINE TABLE REPULSION */
 236             vfitab           = _mm_add_epi32(vfitab,ifour);
 237             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Heps             = _mm_mul_pd(vfeps,H);
 244             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 245             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 246             vvdw12           = _mm_mul_pd(c12_00,VV);
 247             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 248             fvdw12           = _mm_mul_pd(c12_00,FF);
 249             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 250             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm_add_pd(velecsum,velec);
 254             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 255
 256             fscal            = _mm_add_pd(felec,fvdw);
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_pd(fscal,dx00);
 260             ty               = _mm_mul_pd(fscal,dy00);
 261             tz               = _mm_mul_pd(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_add_pd(fix0,tx);
 265             fiy0             = _mm_add_pd(fiy0,ty);
 266             fiz0             = _mm_add_pd(fiz0,tz);
 267
 268             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 269
 270             /* Inner loop uses 73 flops */
 271         }
 272
 273         if(jidx<j_index_end)
 274         {
 275
 276             jnrA             = jjnr[jidx];
 277             j_coord_offsetA  = DIM*jnrA;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_pd(ix0,jx0);
 285             dy00             = _mm_sub_pd(iy0,jy0);
 286             dz00             = _mm_sub_pd(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 292
 293             /* Load parameters for j particles */
 294             jq0              = _mm_load_sd(charge+jnrA+0);
 295             vdwjidx0A        = 2*vdwtype[jnrA+0];
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r00              = _mm_mul_pd(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_pd(iq0,jq0);
 305             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_pd(r00,vftabscale);
 309             vfitab           = _mm_cvttpd_epi32(rt);
 310             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 311             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 312
 313             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm_mul_pd(qq00,VV);
 324             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 325             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 326
 327             /* CUBIC SPLINE TABLE DISPERSION */
 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             vvdw6            = _mm_mul_pd(c6_00,VV);
 339             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 340             fvdw6            = _mm_mul_pd(c6_00,FF);
 341
 342             /* CUBIC SPLINE TABLE REPULSION */
 343             vfitab           = _mm_add_epi32(vfitab,ifour);
 344             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 345             F                = _mm_setzero_pd();
 346             GMX_MM_TRANSPOSE2_PD(Y,F);
 347             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 348             H                = _mm_setzero_pd();
 349             GMX_MM_TRANSPOSE2_PD(G,H);
 350             Heps             = _mm_mul_pd(vfeps,H);
 351             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 352             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 353             vvdw12           = _mm_mul_pd(c12_00,VV);
 354             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 355             fvdw12           = _mm_mul_pd(c12_00,FF);
 356             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 357             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 361             velecsum         = _mm_add_pd(velecsum,velec);
 362             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 363             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 364
 365             fscal            = _mm_add_pd(felec,fvdw);
 366
 367             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm_mul_pd(fscal,dx00);
 371             ty               = _mm_mul_pd(fscal,dy00);
 372             tz               = _mm_mul_pd(fscal,dz00);
 373
 374             /* Update vectorial force */
 375             fix0             = _mm_add_pd(fix0,tx);
 376             fiy0             = _mm_add_pd(fiy0,ty);
 377             fiz0             = _mm_add_pd(fiz0,tz);
 378
 379             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 380
 381             /* Inner loop uses 73 flops */
 382         }
 383
 384         /* End of innermost loop */
 385
 386         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 387                                               f+i_coord_offset,fshift+i_shift_offset);
 388
 389         ggid                        = gid[iidx];
 390         /* Update potential energies */
 391         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 392         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 393
 394         /* Increment number of inner iterations */
 395         inneriter                  += j_index_end - j_index_start;
 396
 397         /* Outer loop uses 9 flops */
 398     }
 399
 400     /* Increment number of outer iterations */
 401     outeriter        += nri;
 402
 403     /* Update outer/inner flops */
 404
 405     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 406 }
 407 /*
 408  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_double
 409  * Electrostatics interaction: CubicSplineTable
 410  * VdW interaction:            CubicSplineTable
 411  * Geometry:                   Particle-Particle
 412  * Calculate force/pot:        Force
 413  */
 414 void
 415 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_double
 416                     (t_nblist                    * gmx_restrict       nlist,
 417                      rvec                        * gmx_restrict          xx,
 418                      rvec                        * gmx_restrict          ff,
 419                      t_forcerec                  * gmx_restrict          fr,
 420                      t_mdatoms                   * gmx_restrict     mdatoms,
 421                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 422                      t_nrnb                      * gmx_restrict        nrnb)
 423 {
 424     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 425      * just 0 for non-waters.
 426      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 427      * jnr indices corresponding to data put in the four positions in the SIMD register.
 428      */
 429     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 430     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 431     int              jnrA,jnrB;
 432     int              j_coord_offsetA,j_coord_offsetB;
 433     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 434     real             rcutoff_scalar;
 435     real             *shiftvec,*fshift,*x,*f;
 436     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 437     int              vdwioffset0;
 438     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 439     int              vdwjidx0A,vdwjidx0B;
 440     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 441     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 442     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 443     real             *charge;
 444     int              nvdwtype;
 445     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 446     int              *vdwtype;
 447     real             *vdwparam;
 448     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 449     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 450     __m128i          vfitab;
 451     __m128i          ifour       = _mm_set1_epi32(4);
 452     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 453     real             *vftab;
 454     __m128d          dummy_mask,cutoff_mask;
 455     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 456     __m128d          one     = _mm_set1_pd(1.0);
 457     __m128d          two     = _mm_set1_pd(2.0);
 458     x                = xx[0];
 459     f                = ff[0];
 460
 461     nri              = nlist->nri;
 462     iinr             = nlist->iinr;
 463     jindex           = nlist->jindex;
 464     jjnr             = nlist->jjnr;
 465     shiftidx         = nlist->shift;
 466     gid              = nlist->gid;
 467     shiftvec         = fr->shift_vec[0];
 468     fshift           = fr->fshift[0];
 469     facel            = _mm_set1_pd(fr->epsfac);
 470     charge           = mdatoms->chargeA;
 471     nvdwtype         = fr->ntype;
 472     vdwparam         = fr->nbfp;
 473     vdwtype          = mdatoms->typeA;
 474
 475     vftab            = kernel_data->table_elec_vdw->data;
 476     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 477
 478     /* Avoid stupid compiler warnings */
 479     jnrA = jnrB = 0;
 480     j_coord_offsetA = 0;
 481     j_coord_offsetB = 0;
 482
 483     outeriter        = 0;
 484     inneriter        = 0;
 485
 486     /* Start outer loop over neighborlists */
 487     for(iidx=0; iidx<nri; iidx++)
 488     {
 489         /* Load shift vector for this list */
 490         i_shift_offset   = DIM*shiftidx[iidx];
 491
 492         /* Load limits for loop over neighbors */
 493         j_index_start    = jindex[iidx];
 494         j_index_end      = jindex[iidx+1];
 495
 496         /* Get outer coordinate index */
 497         inr              = iinr[iidx];
 498         i_coord_offset   = DIM*inr;
 499
 500         /* Load i particle coords and add shift vector */
 501         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 502
 503         fix0             = _mm_setzero_pd();
 504         fiy0             = _mm_setzero_pd();
 505         fiz0             = _mm_setzero_pd();
 506
 507         /* Load parameters for i particles */
 508         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 509         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 510
 511         /* Start inner kernel loop */
 512         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 513         {
 514
 515             /* Get j neighbor index, and coordinate index */
 516             jnrA             = jjnr[jidx];
 517             jnrB             = jjnr[jidx+1];
 518             j_coord_offsetA  = DIM*jnrA;
 519             j_coord_offsetB  = DIM*jnrB;
 520
 521             /* load j atom coordinates */
 522             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 523                                               &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm_sub_pd(ix0,jx0);
 527             dy00             = _mm_sub_pd(iy0,jy0);
 528             dz00             = _mm_sub_pd(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 532
 533             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 537             vdwjidx0A        = 2*vdwtype[jnrA+0];
 538             vdwjidx0B        = 2*vdwtype[jnrB+0];
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r00              = _mm_mul_pd(rsq00,rinv00);
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq00             = _mm_mul_pd(iq0,jq0);
 548             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 549                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 550
 551             /* Calculate table index by multiplying r with table scale and truncate to integer */
 552             rt               = _mm_mul_pd(r00,vftabscale);
 553             vfitab           = _mm_cvttpd_epi32(rt);
 554             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 555             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 556
 557             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 558             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 559             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 560             GMX_MM_TRANSPOSE2_PD(Y,F);
 561             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 562             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 563             GMX_MM_TRANSPOSE2_PD(G,H);
 564             Heps             = _mm_mul_pd(vfeps,H);
 565             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 566             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 567             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 568
 569             /* CUBIC SPLINE TABLE DISPERSION */
 570             vfitab           = _mm_add_epi32(vfitab,ifour);
 571             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 572             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 573             GMX_MM_TRANSPOSE2_PD(Y,F);
 574             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 575             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 576             GMX_MM_TRANSPOSE2_PD(G,H);
 577             Heps             = _mm_mul_pd(vfeps,H);
 578             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 579             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 580             fvdw6            = _mm_mul_pd(c6_00,FF);
 581
 582             /* CUBIC SPLINE TABLE REPULSION */
 583             vfitab           = _mm_add_epi32(vfitab,ifour);
 584             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 585             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 586             GMX_MM_TRANSPOSE2_PD(Y,F);
 587             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 588             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 589             GMX_MM_TRANSPOSE2_PD(G,H);
 590             Heps             = _mm_mul_pd(vfeps,H);
 591             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 592             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 593             fvdw12           = _mm_mul_pd(c12_00,FF);
 594             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 595
 596             fscal            = _mm_add_pd(felec,fvdw);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_pd(fscal,dx00);
 600             ty               = _mm_mul_pd(fscal,dy00);
 601             tz               = _mm_mul_pd(fscal,dz00);
 602
 603             /* Update vectorial force */
 604             fix0             = _mm_add_pd(fix0,tx);
 605             fiy0             = _mm_add_pd(fiy0,ty);
 606             fiz0             = _mm_add_pd(fiz0,tz);
 607
 608             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 609
 610             /* Inner loop uses 61 flops */
 611         }
 612
 613         if(jidx<j_index_end)
 614         {
 615
 616             jnrA             = jjnr[jidx];
 617             j_coord_offsetA  = DIM*jnrA;
 618
 619             /* load j atom coordinates */
 620             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 621                                               &jx0,&jy0,&jz0);
 622
 623             /* Calculate displacement vector */
 624             dx00             = _mm_sub_pd(ix0,jx0);
 625             dy00             = _mm_sub_pd(iy0,jy0);
 626             dz00             = _mm_sub_pd(iz0,jz0);
 627
 628             /* Calculate squared distance and things based on it */
 629             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 630
 631             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 632
 633             /* Load parameters for j particles */
 634             jq0              = _mm_load_sd(charge+jnrA+0);
 635             vdwjidx0A        = 2*vdwtype[jnrA+0];
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             r00              = _mm_mul_pd(rsq00,rinv00);
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             qq00             = _mm_mul_pd(iq0,jq0);
 645             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 646
 647             /* Calculate table index by multiplying r with table scale and truncate to integer */
 648             rt               = _mm_mul_pd(r00,vftabscale);
 649             vfitab           = _mm_cvttpd_epi32(rt);
 650             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 651             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 652
 653             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 654             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 655             F                = _mm_setzero_pd();
 656             GMX_MM_TRANSPOSE2_PD(Y,F);
 657             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 658             H                = _mm_setzero_pd();
 659             GMX_MM_TRANSPOSE2_PD(G,H);
 660             Heps             = _mm_mul_pd(vfeps,H);
 661             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 662             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 663             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 664
 665             /* CUBIC SPLINE TABLE DISPERSION */
 666             vfitab           = _mm_add_epi32(vfitab,ifour);
 667             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 668             F                = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(Y,F);
 670             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 671             H                = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(G,H);
 673             Heps             = _mm_mul_pd(vfeps,H);
 674             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 675             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 676             fvdw6            = _mm_mul_pd(c6_00,FF);
 677
 678             /* CUBIC SPLINE TABLE REPULSION */
 679             vfitab           = _mm_add_epi32(vfitab,ifour);
 680             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 681             F                = _mm_setzero_pd();
 682             GMX_MM_TRANSPOSE2_PD(Y,F);
 683             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 684             H                = _mm_setzero_pd();
 685             GMX_MM_TRANSPOSE2_PD(G,H);
 686             Heps             = _mm_mul_pd(vfeps,H);
 687             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 688             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 689             fvdw12           = _mm_mul_pd(c12_00,FF);
 690             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 691
 692             fscal            = _mm_add_pd(felec,fvdw);
 693
 694             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm_mul_pd(fscal,dx00);
 698             ty               = _mm_mul_pd(fscal,dy00);
 699             tz               = _mm_mul_pd(fscal,dz00);
 700
 701             /* Update vectorial force */
 702             fix0             = _mm_add_pd(fix0,tx);
 703             fiy0             = _mm_add_pd(fiy0,ty);
 704             fiz0             = _mm_add_pd(fiz0,tz);
 705
 706             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 707
 708             /* Inner loop uses 61 flops */
 709         }
 710
 711         /* End of innermost loop */
 712
 713         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 714                                               f+i_coord_offset,fshift+i_shift_offset);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 7 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 728 }