src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecNone_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse2_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     int              nvdwtype;
  86     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  90     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     vftab            = kernel_data->table_vdw->data;
 115     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149         /* Reset potential sums */
 150         vvdwsum          = _mm_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_pd(ix0,jx0);
 168             dy00             = _mm_sub_pd(iy0,jy0);
 169             dz00             = _mm_sub_pd(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 175
 176             /* Load parameters for j particles */
 177             vdwjidx0A        = 2*vdwtype[jnrA+0];
 178             vdwjidx0B        = 2*vdwtype[jnrB+0];
 179
 180             /**************************
 181              * CALCULATE INTERACTIONS *
 182              **************************/
 183
 184             r00              = _mm_mul_pd(rsq00,rinv00);
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 188                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 189
 190             /* Calculate table index by multiplying r with table scale and truncate to integer */
 191             rt               = _mm_mul_pd(r00,vftabscale);
 192             vfitab           = _mm_cvttpd_epi32(rt);
 193             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 194             vfitab           = _mm_slli_epi32(vfitab,3);
 195
 196             /* CUBIC SPLINE TABLE DISPERSION */
 197             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 198             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 199             GMX_MM_TRANSPOSE2_PD(Y,F);
 200             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 201             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 202             GMX_MM_TRANSPOSE2_PD(G,H);
 203             Heps             = _mm_mul_pd(vfeps,H);
 204             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 205             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 206             vvdw6            = _mm_mul_pd(c6_00,VV);
 207             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 208             fvdw6            = _mm_mul_pd(c6_00,FF);
 209
 210             /* CUBIC SPLINE TABLE REPULSION */
 211             vfitab           = _mm_add_epi32(vfitab,ifour);
 212             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 213             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 214             GMX_MM_TRANSPOSE2_PD(Y,F);
 215             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 216             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 217             GMX_MM_TRANSPOSE2_PD(G,H);
 218             Heps             = _mm_mul_pd(vfeps,H);
 219             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 220             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 221             vvdw12           = _mm_mul_pd(c12_00,VV);
 222             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 223             fvdw12           = _mm_mul_pd(c12_00,FF);
 224             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 225             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_pd(fscal,dx00);
 234             ty               = _mm_mul_pd(fscal,dy00);
 235             tz               = _mm_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_pd(fix0,tx);
 239             fiy0             = _mm_add_pd(fiy0,ty);
 240             fiz0             = _mm_add_pd(fiz0,tz);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 243
 244             /* Inner loop uses 56 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_pd(ix0,jx0);
 259             dy00             = _mm_sub_pd(iy0,jy0);
 260             dz00             = _mm_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 266
 267             /* Load parameters for j particles */
 268             vdwjidx0A        = 2*vdwtype[jnrA+0];
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_pd(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 278
 279             /* Calculate table index by multiplying r with table scale and truncate to integer */
 280             rt               = _mm_mul_pd(r00,vftabscale);
 281             vfitab           = _mm_cvttpd_epi32(rt);
 282             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 283             vfitab           = _mm_slli_epi32(vfitab,3);
 284
 285             /* CUBIC SPLINE TABLE DISPERSION */
 286             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 287             F                = _mm_setzero_pd();
 288             GMX_MM_TRANSPOSE2_PD(Y,F);
 289             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 290             H                = _mm_setzero_pd();
 291             GMX_MM_TRANSPOSE2_PD(G,H);
 292             Heps             = _mm_mul_pd(vfeps,H);
 293             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 294             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 295             vvdw6            = _mm_mul_pd(c6_00,VV);
 296             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 297             fvdw6            = _mm_mul_pd(c6_00,FF);
 298
 299             /* CUBIC SPLINE TABLE REPULSION */
 300             vfitab           = _mm_add_epi32(vfitab,ifour);
 301             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 302             F                = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(Y,F);
 304             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 305             H                = _mm_setzero_pd();
 306             GMX_MM_TRANSPOSE2_PD(G,H);
 307             Heps             = _mm_mul_pd(vfeps,H);
 308             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 309             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 310             vvdw12           = _mm_mul_pd(c12_00,VV);
 311             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 312             fvdw12           = _mm_mul_pd(c12_00,FF);
 313             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 314             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 318             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 319
 320             fscal            = fvdw;
 321
 322             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_pd(fscal,dx00);
 326             ty               = _mm_mul_pd(fscal,dy00);
 327             tz               = _mm_mul_pd(fscal,dz00);
 328
 329             /* Update vectorial force */
 330             fix0             = _mm_add_pd(fix0,tx);
 331             fiy0             = _mm_add_pd(fiy0,ty);
 332             fiz0             = _mm_add_pd(fiz0,tz);
 333
 334             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 335
 336             /* Inner loop uses 56 flops */
 337         }
 338
 339         /* End of innermost loop */
 340
 341         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 342                                               f+i_coord_offset,fshift+i_shift_offset);
 343
 344         ggid                        = gid[iidx];
 345         /* Update potential energies */
 346         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 347
 348         /* Increment number of inner iterations */
 349         inneriter                  += j_index_end - j_index_start;
 350
 351         /* Outer loop uses 7 flops */
 352     }
 353
 354     /* Increment number of outer iterations */
 355     outeriter        += nri;
 356
 357     /* Update outer/inner flops */
 358
 359     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*56);
 360 }
 361 /*
 362  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse2_double
 363  * Electrostatics interaction: None
 364  * VdW interaction:            CubicSplineTable
 365  * Geometry:                   Particle-Particle
 366  * Calculate force/pot:        Force
 367  */
 368 void
 369 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse2_double
 370                     (t_nblist                    * gmx_restrict       nlist,
 371                      rvec                        * gmx_restrict          xx,
 372                      rvec                        * gmx_restrict          ff,
 373                      t_forcerec                  * gmx_restrict          fr,
 374                      t_mdatoms                   * gmx_restrict     mdatoms,
 375                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 376                      t_nrnb                      * gmx_restrict        nrnb)
 377 {
 378     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 379      * just 0 for non-waters.
 380      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 381      * jnr indices corresponding to data put in the four positions in the SIMD register.
 382      */
 383     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 384     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 385     int              jnrA,jnrB;
 386     int              j_coord_offsetA,j_coord_offsetB;
 387     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 388     real             rcutoff_scalar;
 389     real             *shiftvec,*fshift,*x,*f;
 390     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 391     int              vdwioffset0;
 392     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 393     int              vdwjidx0A,vdwjidx0B;
 394     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 395     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 396     int              nvdwtype;
 397     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 398     int              *vdwtype;
 399     real             *vdwparam;
 400     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 401     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 402     __m128i          vfitab;
 403     __m128i          ifour       = _mm_set1_epi32(4);
 404     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 405     real             *vftab;
 406     __m128d          dummy_mask,cutoff_mask;
 407     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 408     __m128d          one     = _mm_set1_pd(1.0);
 409     __m128d          two     = _mm_set1_pd(2.0);
 410     x                = xx[0];
 411     f                = ff[0];
 412
 413     nri              = nlist->nri;
 414     iinr             = nlist->iinr;
 415     jindex           = nlist->jindex;
 416     jjnr             = nlist->jjnr;
 417     shiftidx         = nlist->shift;
 418     gid              = nlist->gid;
 419     shiftvec         = fr->shift_vec[0];
 420     fshift           = fr->fshift[0];
 421     nvdwtype         = fr->ntype;
 422     vdwparam         = fr->nbfp;
 423     vdwtype          = mdatoms->typeA;
 424
 425     vftab            = kernel_data->table_vdw->data;
 426     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432
 433     outeriter        = 0;
 434     inneriter        = 0;
 435
 436     /* Start outer loop over neighborlists */
 437     for(iidx=0; iidx<nri; iidx++)
 438     {
 439         /* Load shift vector for this list */
 440         i_shift_offset   = DIM*shiftidx[iidx];
 441
 442         /* Load limits for loop over neighbors */
 443         j_index_start    = jindex[iidx];
 444         j_index_end      = jindex[iidx+1];
 445
 446         /* Get outer coordinate index */
 447         inr              = iinr[iidx];
 448         i_coord_offset   = DIM*inr;
 449
 450         /* Load i particle coords and add shift vector */
 451         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 452
 453         fix0             = _mm_setzero_pd();
 454         fiy0             = _mm_setzero_pd();
 455         fiz0             = _mm_setzero_pd();
 456
 457         /* Load parameters for i particles */
 458         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 459
 460         /* Start inner kernel loop */
 461         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrA             = jjnr[jidx];
 466             jnrB             = jjnr[jidx+1];
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469
 470             /* load j atom coordinates */
 471             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm_sub_pd(ix0,jx0);
 476             dy00             = _mm_sub_pd(iy0,jy0);
 477             dz00             = _mm_sub_pd(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 481
 482             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 483
 484             /* Load parameters for j particles */
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r00              = _mm_mul_pd(rsq00,rinv00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 496                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 497
 498             /* Calculate table index by multiplying r with table scale and truncate to integer */
 499             rt               = _mm_mul_pd(r00,vftabscale);
 500             vfitab           = _mm_cvttpd_epi32(rt);
 501             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 502             vfitab           = _mm_slli_epi32(vfitab,3);
 503
 504             /* CUBIC SPLINE TABLE DISPERSION */
 505             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 506             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 507             GMX_MM_TRANSPOSE2_PD(Y,F);
 508             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 509             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 510             GMX_MM_TRANSPOSE2_PD(G,H);
 511             Heps             = _mm_mul_pd(vfeps,H);
 512             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 513             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 514             fvdw6            = _mm_mul_pd(c6_00,FF);
 515
 516             /* CUBIC SPLINE TABLE REPULSION */
 517             vfitab           = _mm_add_epi32(vfitab,ifour);
 518             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 519             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 520             GMX_MM_TRANSPOSE2_PD(Y,F);
 521             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 522             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 523             GMX_MM_TRANSPOSE2_PD(G,H);
 524             Heps             = _mm_mul_pd(vfeps,H);
 525             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 526             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 527             fvdw12           = _mm_mul_pd(c12_00,FF);
 528             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 529
 530             fscal            = fvdw;
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_pd(fscal,dx00);
 534             ty               = _mm_mul_pd(fscal,dy00);
 535             tz               = _mm_mul_pd(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_add_pd(fix0,tx);
 539             fiy0             = _mm_add_pd(fiy0,ty);
 540             fiz0             = _mm_add_pd(fiz0,tz);
 541
 542             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 543
 544             /* Inner loop uses 48 flops */
 545         }
 546
 547         if(jidx<j_index_end)
 548         {
 549
 550             jnrA             = jjnr[jidx];
 551             j_coord_offsetA  = DIM*jnrA;
 552
 553             /* load j atom coordinates */
 554             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_pd(ix0,jx0);
 559             dy00             = _mm_sub_pd(iy0,jy0);
 560             dz00             = _mm_sub_pd(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 564
 565             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 566
 567             /* Load parameters for j particles */
 568             vdwjidx0A        = 2*vdwtype[jnrA+0];
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             r00              = _mm_mul_pd(rsq00,rinv00);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 578
 579             /* Calculate table index by multiplying r with table scale and truncate to integer */
 580             rt               = _mm_mul_pd(r00,vftabscale);
 581             vfitab           = _mm_cvttpd_epi32(rt);
 582             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 583             vfitab           = _mm_slli_epi32(vfitab,3);
 584
 585             /* CUBIC SPLINE TABLE DISPERSION */
 586             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 587             F                = _mm_setzero_pd();
 588             GMX_MM_TRANSPOSE2_PD(Y,F);
 589             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 590             H                = _mm_setzero_pd();
 591             GMX_MM_TRANSPOSE2_PD(G,H);
 592             Heps             = _mm_mul_pd(vfeps,H);
 593             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 594             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 595             fvdw6            = _mm_mul_pd(c6_00,FF);
 596
 597             /* CUBIC SPLINE TABLE REPULSION */
 598             vfitab           = _mm_add_epi32(vfitab,ifour);
 599             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 600             F                = _mm_setzero_pd();
 601             GMX_MM_TRANSPOSE2_PD(Y,F);
 602             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 603             H                = _mm_setzero_pd();
 604             GMX_MM_TRANSPOSE2_PD(G,H);
 605             Heps             = _mm_mul_pd(vfeps,H);
 606             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 607             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 608             fvdw12           = _mm_mul_pd(c12_00,FF);
 609             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 610
 611             fscal            = fvdw;
 612
 613             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx00);
 617             ty               = _mm_mul_pd(fscal,dy00);
 618             tz               = _mm_mul_pd(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_add_pd(fix0,tx);
 622             fiy0             = _mm_add_pd(fiy0,ty);
 623             fiz0             = _mm_add_pd(fiz0,tz);
 624
 625             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 626
 627             /* Inner loop uses 48 flops */
 628         }
 629
 630         /* End of innermost loop */
 631
 632         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 633                                               f+i_coord_offset,fshift+i_shift_offset);
 634
 635         /* Increment number of inner iterations */
 636         inneriter                  += j_index_end - j_index_start;
 637
 638         /* Outer loop uses 6 flops */
 639     }
 640
 641     /* Increment number of outer iterations */
 642     outeriter        += nri;
 643
 644     /* Update outer/inner flops */
 645
 646     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 647 }