src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_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     int              nvdwtype;
  85     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  89     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  90     __m128i          vfitab;
  91     __m128i          ifour       = _mm_set1_epi32(4);
  92     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  93     real             *vftab;
  94     __m128d          dummy_mask,cutoff_mask;
  95     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  96     __m128d          one     = _mm_set1_pd(1.0);
  97     __m128d          two     = _mm_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm_setzero_pd();
 142         fiy0             = _mm_setzero_pd();
 143         fiz0             = _mm_setzero_pd();
 144
 145         /* Load parameters for i particles */
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         vvdwsum          = _mm_setzero_pd();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm_sub_pd(ix0,jx0);
 167             dy00             = _mm_sub_pd(iy0,jy0);
 168             dz00             = _mm_sub_pd(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 172
 173             rinv00           = sse41_invsqrt_d(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             r00              = _mm_mul_pd(rsq00,rinv00);
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 187                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 188
 189             /* Calculate table index by multiplying r with table scale and truncate to integer */
 190             rt               = _mm_mul_pd(r00,vftabscale);
 191             vfitab           = _mm_cvttpd_epi32(rt);
 192             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 193             vfitab           = _mm_slli_epi32(vfitab,3);
 194
 195             /* CUBIC SPLINE TABLE DISPERSION */
 196             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 197             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 198             GMX_MM_TRANSPOSE2_PD(Y,F);
 199             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 200             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 201             GMX_MM_TRANSPOSE2_PD(G,H);
 202             Heps             = _mm_mul_pd(vfeps,H);
 203             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 204             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 205             vvdw6            = _mm_mul_pd(c6_00,VV);
 206             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 207             fvdw6            = _mm_mul_pd(c6_00,FF);
 208
 209             /* CUBIC SPLINE TABLE REPULSION */
 210             vfitab           = _mm_add_epi32(vfitab,ifour);
 211             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 212             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 213             GMX_MM_TRANSPOSE2_PD(Y,F);
 214             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 215             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 216             GMX_MM_TRANSPOSE2_PD(G,H);
 217             Heps             = _mm_mul_pd(vfeps,H);
 218             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 219             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 220             vvdw12           = _mm_mul_pd(c12_00,VV);
 221             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 222             fvdw12           = _mm_mul_pd(c12_00,FF);
 223             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 224             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 228
 229             fscal            = fvdw;
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm_mul_pd(fscal,dx00);
 233             ty               = _mm_mul_pd(fscal,dy00);
 234             tz               = _mm_mul_pd(fscal,dz00);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm_add_pd(fix0,tx);
 238             fiy0             = _mm_add_pd(fiy0,ty);
 239             fiz0             = _mm_add_pd(fiz0,tz);
 240
 241             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 242
 243             /* Inner loop uses 56 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             jnrA             = jjnr[jidx];
 250             j_coord_offsetA  = DIM*jnrA;
 251
 252             /* load j atom coordinates */
 253             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm_sub_pd(ix0,jx0);
 258             dy00             = _mm_sub_pd(iy0,jy0);
 259             dz00             = _mm_sub_pd(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 263
 264             rinv00           = sse41_invsqrt_d(rsq00);
 265
 266             /* Load parameters for j particles */
 267             vdwjidx0A        = 2*vdwtype[jnrA+0];
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r00              = _mm_mul_pd(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm_mul_pd(r00,vftabscale);
 280             vfitab           = _mm_cvttpd_epi32(rt);
 281             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 282             vfitab           = _mm_slli_epi32(vfitab,3);
 283
 284             /* CUBIC SPLINE TABLE DISPERSION */
 285             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 286             F                = _mm_setzero_pd();
 287             GMX_MM_TRANSPOSE2_PD(Y,F);
 288             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 289             H                = _mm_setzero_pd();
 290             GMX_MM_TRANSPOSE2_PD(G,H);
 291             Heps             = _mm_mul_pd(vfeps,H);
 292             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 293             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 294             vvdw6            = _mm_mul_pd(c6_00,VV);
 295             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 296             fvdw6            = _mm_mul_pd(c6_00,FF);
 297
 298             /* CUBIC SPLINE TABLE REPULSION */
 299             vfitab           = _mm_add_epi32(vfitab,ifour);
 300             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 301             F                = _mm_setzero_pd();
 302             GMX_MM_TRANSPOSE2_PD(Y,F);
 303             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 304             H                = _mm_setzero_pd();
 305             GMX_MM_TRANSPOSE2_PD(G,H);
 306             Heps             = _mm_mul_pd(vfeps,H);
 307             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 308             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 309             vvdw12           = _mm_mul_pd(c12_00,VV);
 310             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 311             fvdw12           = _mm_mul_pd(c12_00,FF);
 312             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 313             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 317             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 318
 319             fscal            = fvdw;
 320
 321             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_pd(fscal,dx00);
 325             ty               = _mm_mul_pd(fscal,dy00);
 326             tz               = _mm_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_add_pd(fix0,tx);
 330             fiy0             = _mm_add_pd(fiy0,ty);
 331             fiz0             = _mm_add_pd(fiz0,tz);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 334
 335             /* Inner loop uses 56 flops */
 336         }
 337
 338         /* End of innermost loop */
 339
 340         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 341                                               f+i_coord_offset,fshift+i_shift_offset);
 342
 343         ggid                        = gid[iidx];
 344         /* Update potential energies */
 345         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 346
 347         /* Increment number of inner iterations */
 348         inneriter                  += j_index_end - j_index_start;
 349
 350         /* Outer loop uses 7 flops */
 351     }
 352
 353     /* Increment number of outer iterations */
 354     outeriter        += nri;
 355
 356     /* Update outer/inner flops */
 357
 358     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*56);
 359 }
 360 /*
 361  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 362  * Electrostatics interaction: None
 363  * VdW interaction:            CubicSplineTable
 364  * Geometry:                   Particle-Particle
 365  * Calculate force/pot:        Force
 366  */
 367 void
 368 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 369                     (t_nblist                    * gmx_restrict       nlist,
 370                      rvec                        * gmx_restrict          xx,
 371                      rvec                        * gmx_restrict          ff,
 372                      struct t_forcerec           * gmx_restrict          fr,
 373                      t_mdatoms                   * gmx_restrict     mdatoms,
 374                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 375                      t_nrnb                      * gmx_restrict        nrnb)
 376 {
 377     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 378      * just 0 for non-waters.
 379      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 380      * jnr indices corresponding to data put in the four positions in the SIMD register.
 381      */
 382     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 383     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 384     int              jnrA,jnrB;
 385     int              j_coord_offsetA,j_coord_offsetB;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 390     int              vdwioffset0;
 391     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B;
 393     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     int              nvdwtype;
 396     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 397     int              *vdwtype;
 398     real             *vdwparam;
 399     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 400     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 401     __m128i          vfitab;
 402     __m128i          ifour       = _mm_set1_epi32(4);
 403     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 404     real             *vftab;
 405     __m128d          dummy_mask,cutoff_mask;
 406     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 407     __m128d          one     = _mm_set1_pd(1.0);
 408     __m128d          two     = _mm_set1_pd(2.0);
 409     x                = xx[0];
 410     f                = ff[0];
 411
 412     nri              = nlist->nri;
 413     iinr             = nlist->iinr;
 414     jindex           = nlist->jindex;
 415     jjnr             = nlist->jjnr;
 416     shiftidx         = nlist->shift;
 417     gid              = nlist->gid;
 418     shiftvec         = fr->shift_vec[0];
 419     fshift           = fr->fshift[0];
 420     nvdwtype         = fr->ntype;
 421     vdwparam         = fr->nbfp;
 422     vdwtype          = mdatoms->typeA;
 423
 424     vftab            = kernel_data->table_vdw->data;
 425     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     /* Start outer loop over neighborlists */
 436     for(iidx=0; iidx<nri; iidx++)
 437     {
 438         /* Load shift vector for this list */
 439         i_shift_offset   = DIM*shiftidx[iidx];
 440
 441         /* Load limits for loop over neighbors */
 442         j_index_start    = jindex[iidx];
 443         j_index_end      = jindex[iidx+1];
 444
 445         /* Get outer coordinate index */
 446         inr              = iinr[iidx];
 447         i_coord_offset   = DIM*inr;
 448
 449         /* Load i particle coords and add shift vector */
 450         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 451
 452         fix0             = _mm_setzero_pd();
 453         fiy0             = _mm_setzero_pd();
 454         fiz0             = _mm_setzero_pd();
 455
 456         /* Load parameters for i particles */
 457         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 458
 459         /* Start inner kernel loop */
 460         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrA             = jjnr[jidx];
 465             jnrB             = jjnr[jidx+1];
 466             j_coord_offsetA  = DIM*jnrA;
 467             j_coord_offsetB  = DIM*jnrB;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_pd(ix0,jx0);
 475             dy00             = _mm_sub_pd(iy0,jy0);
 476             dz00             = _mm_sub_pd(iz0,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 480
 481             rinv00           = sse41_invsqrt_d(rsq00);
 482
 483             /* Load parameters for j particles */
 484             vdwjidx0A        = 2*vdwtype[jnrA+0];
 485             vdwjidx0B        = 2*vdwtype[jnrB+0];
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             r00              = _mm_mul_pd(rsq00,rinv00);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 495                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 496
 497             /* Calculate table index by multiplying r with table scale and truncate to integer */
 498             rt               = _mm_mul_pd(r00,vftabscale);
 499             vfitab           = _mm_cvttpd_epi32(rt);
 500             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 501             vfitab           = _mm_slli_epi32(vfitab,3);
 502
 503             /* CUBIC SPLINE TABLE DISPERSION */
 504             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 505             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 506             GMX_MM_TRANSPOSE2_PD(Y,F);
 507             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 508             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 509             GMX_MM_TRANSPOSE2_PD(G,H);
 510             Heps             = _mm_mul_pd(vfeps,H);
 511             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 512             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 513             fvdw6            = _mm_mul_pd(c6_00,FF);
 514
 515             /* CUBIC SPLINE TABLE REPULSION */
 516             vfitab           = _mm_add_epi32(vfitab,ifour);
 517             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 518             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 519             GMX_MM_TRANSPOSE2_PD(Y,F);
 520             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 521             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 522             GMX_MM_TRANSPOSE2_PD(G,H);
 523             Heps             = _mm_mul_pd(vfeps,H);
 524             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 525             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 526             fvdw12           = _mm_mul_pd(c12_00,FF);
 527             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 528
 529             fscal            = fvdw;
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 542
 543             /* Inner loop uses 48 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             jnrA             = jjnr[jidx];
 550             j_coord_offsetA  = DIM*jnrA;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx00             = _mm_sub_pd(ix0,jx0);
 558             dy00             = _mm_sub_pd(iy0,jy0);
 559             dz00             = _mm_sub_pd(iz0,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 563
 564             rinv00           = sse41_invsqrt_d(rsq00);
 565
 566             /* Load parameters for j particles */
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r00              = _mm_mul_pd(rsq00,rinv00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 577
 578             /* Calculate table index by multiplying r with table scale and truncate to integer */
 579             rt               = _mm_mul_pd(r00,vftabscale);
 580             vfitab           = _mm_cvttpd_epi32(rt);
 581             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 582             vfitab           = _mm_slli_epi32(vfitab,3);
 583
 584             /* CUBIC SPLINE TABLE DISPERSION */
 585             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 586             F                = _mm_setzero_pd();
 587             GMX_MM_TRANSPOSE2_PD(Y,F);
 588             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 589             H                = _mm_setzero_pd();
 590             GMX_MM_TRANSPOSE2_PD(G,H);
 591             Heps             = _mm_mul_pd(vfeps,H);
 592             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 593             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 594             fvdw6            = _mm_mul_pd(c6_00,FF);
 595
 596             /* CUBIC SPLINE TABLE REPULSION */
 597             vfitab           = _mm_add_epi32(vfitab,ifour);
 598             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 599             F                = _mm_setzero_pd();
 600             GMX_MM_TRANSPOSE2_PD(Y,F);
 601             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 602             H                = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(G,H);
 604             Heps             = _mm_mul_pd(vfeps,H);
 605             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 606             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 607             fvdw12           = _mm_mul_pd(c12_00,FF);
 608             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 609
 610             fscal            = fvdw;
 611
 612             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm_mul_pd(fscal,dx00);
 616             ty               = _mm_mul_pd(fscal,dy00);
 617             tz               = _mm_mul_pd(fscal,dz00);
 618
 619             /* Update vectorial force */
 620             fix0             = _mm_add_pd(fix0,tx);
 621             fiy0             = _mm_add_pd(fiy0,ty);
 622             fiz0             = _mm_add_pd(fiz0,tz);
 623
 624             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 625
 626             /* Inner loop uses 48 flops */
 627         }
 628
 629         /* End of innermost loop */
 630
 631         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 632                                               f+i_coord_offset,fshift+i_shift_offset);
 633
 634         /* Increment number of inner iterations */
 635         inneriter                  += j_index_end - j_index_start;
 636
 637         /* Outer loop uses 6 flops */
 638     }
 639
 640     /* Increment number of outer iterations */
 641     outeriter        += nri;
 642
 643     /* Update outer/inner flops */
 644
 645     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 646 }