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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: None
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_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     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_pd();
 145         fiy0             = _mm_setzero_pd();
 146         fiz0             = _mm_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         vvdwsum          = _mm_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_pd(ix0,jx0);
 170             dy00             = _mm_sub_pd(iy0,jy0);
 171             dz00             = _mm_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 177
 178             /* Load parameters for j particles */
 179             vdwjidx0A        = 2*vdwtype[jnrA+0];
 180             vdwjidx0B        = 2*vdwtype[jnrB+0];
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             r00              = _mm_mul_pd(rsq00,rinv00);
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 190                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 191
 192             /* Calculate table index by multiplying r with table scale and truncate to integer */
 193             rt               = _mm_mul_pd(r00,vftabscale);
 194             vfitab           = _mm_cvttpd_epi32(rt);
 195             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 196             vfitab           = _mm_slli_epi32(vfitab,3);
 197
 198             /* CUBIC SPLINE TABLE DISPERSION */
 199             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 200             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 201             GMX_MM_TRANSPOSE2_PD(Y,F);
 202             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 203             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 204             GMX_MM_TRANSPOSE2_PD(G,H);
 205             Heps             = _mm_mul_pd(vfeps,H);
 206             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 207             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 208             vvdw6            = _mm_mul_pd(c6_00,VV);
 209             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 210             fvdw6            = _mm_mul_pd(c6_00,FF);
 211
 212             /* CUBIC SPLINE TABLE REPULSION */
 213             vfitab           = _mm_add_epi32(vfitab,ifour);
 214             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 215             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 216             GMX_MM_TRANSPOSE2_PD(Y,F);
 217             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 218             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 219             GMX_MM_TRANSPOSE2_PD(G,H);
 220             Heps             = _mm_mul_pd(vfeps,H);
 221             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 222             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 223             vvdw12           = _mm_mul_pd(c12_00,VV);
 224             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 225             fvdw12           = _mm_mul_pd(c12_00,FF);
 226             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 227             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 231
 232             fscal            = fvdw;
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_pd(fscal,dx00);
 236             ty               = _mm_mul_pd(fscal,dy00);
 237             tz               = _mm_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_pd(fix0,tx);
 241             fiy0             = _mm_add_pd(fiy0,ty);
 242             fiz0             = _mm_add_pd(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 245
 246             /* Inner loop uses 56 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 268
 269             /* Load parameters for j particles */
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             r00              = _mm_mul_pd(rsq00,rinv00);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 280
 281             /* Calculate table index by multiplying r with table scale and truncate to integer */
 282             rt               = _mm_mul_pd(r00,vftabscale);
 283             vfitab           = _mm_cvttpd_epi32(rt);
 284             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 285             vfitab           = _mm_slli_epi32(vfitab,3);
 286
 287             /* CUBIC SPLINE TABLE DISPERSION */
 288             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 289             F                = _mm_setzero_pd();
 290             GMX_MM_TRANSPOSE2_PD(Y,F);
 291             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 292             H                = _mm_setzero_pd();
 293             GMX_MM_TRANSPOSE2_PD(G,H);
 294             Heps             = _mm_mul_pd(vfeps,H);
 295             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 296             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 297             vvdw6            = _mm_mul_pd(c6_00,VV);
 298             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 299             fvdw6            = _mm_mul_pd(c6_00,FF);
 300
 301             /* CUBIC SPLINE TABLE REPULSION */
 302             vfitab           = _mm_add_epi32(vfitab,ifour);
 303             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 304             F                = _mm_setzero_pd();
 305             GMX_MM_TRANSPOSE2_PD(Y,F);
 306             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 307             H                = _mm_setzero_pd();
 308             GMX_MM_TRANSPOSE2_PD(G,H);
 309             Heps             = _mm_mul_pd(vfeps,H);
 310             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 311             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 312             vvdw12           = _mm_mul_pd(c12_00,VV);
 313             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 314             fvdw12           = _mm_mul_pd(c12_00,FF);
 315             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 316             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 320             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 321
 322             fscal            = fvdw;
 323
 324             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_pd(fscal,dx00);
 328             ty               = _mm_mul_pd(fscal,dy00);
 329             tz               = _mm_mul_pd(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm_add_pd(fix0,tx);
 333             fiy0             = _mm_add_pd(fiy0,ty);
 334             fiz0             = _mm_add_pd(fiz0,tz);
 335
 336             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 337
 338             /* Inner loop uses 56 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 7 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*56);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 365  * Electrostatics interaction: None
 366  * VdW interaction:            CubicSplineTable
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      t_forcerec                  * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB;
 388     int              j_coord_offsetA,j_coord_offsetB;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     int              vdwioffset0;
 394     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B;
 396     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     int              nvdwtype;
 399     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 400     int              *vdwtype;
 401     real             *vdwparam;
 402     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 403     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 404     __m128i          vfitab;
 405     __m128i          ifour       = _mm_set1_epi32(4);
 406     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 407     real             *vftab;
 408     __m128d          dummy_mask,cutoff_mask;
 409     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 410     __m128d          one     = _mm_set1_pd(1.0);
 411     __m128d          two     = _mm_set1_pd(2.0);
 412     x                = xx[0];
 413     f                = ff[0];
 414
 415     nri              = nlist->nri;
 416     iinr             = nlist->iinr;
 417     jindex           = nlist->jindex;
 418     jjnr             = nlist->jjnr;
 419     shiftidx         = nlist->shift;
 420     gid              = nlist->gid;
 421     shiftvec         = fr->shift_vec[0];
 422     fshift           = fr->fshift[0];
 423     nvdwtype         = fr->ntype;
 424     vdwparam         = fr->nbfp;
 425     vdwtype          = mdatoms->typeA;
 426
 427     vftab            = kernel_data->table_vdw->data;
 428     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _mm_setzero_pd();
 456         fiy0             = _mm_setzero_pd();
 457         fiz0             = _mm_setzero_pd();
 458
 459         /* Load parameters for i particles */
 460         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 461
 462         /* Start inner kernel loop */
 463         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrA             = jjnr[jidx];
 468             jnrB             = jjnr[jidx+1];
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471
 472             /* load j atom coordinates */
 473             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _mm_sub_pd(ix0,jx0);
 478             dy00             = _mm_sub_pd(iy0,jy0);
 479             dz00             = _mm_sub_pd(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483
 484             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 485
 486             /* Load parameters for j particles */
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r00              = _mm_mul_pd(rsq00,rinv00);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 498                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 499
 500             /* Calculate table index by multiplying r with table scale and truncate to integer */
 501             rt               = _mm_mul_pd(r00,vftabscale);
 502             vfitab           = _mm_cvttpd_epi32(rt);
 503             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 504             vfitab           = _mm_slli_epi32(vfitab,3);
 505
 506             /* CUBIC SPLINE TABLE DISPERSION */
 507             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 508             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 509             GMX_MM_TRANSPOSE2_PD(Y,F);
 510             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 511             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 512             GMX_MM_TRANSPOSE2_PD(G,H);
 513             Heps             = _mm_mul_pd(vfeps,H);
 514             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 515             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 516             fvdw6            = _mm_mul_pd(c6_00,FF);
 517
 518             /* CUBIC SPLINE TABLE REPULSION */
 519             vfitab           = _mm_add_epi32(vfitab,ifour);
 520             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 521             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 522             GMX_MM_TRANSPOSE2_PD(Y,F);
 523             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 524             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 525             GMX_MM_TRANSPOSE2_PD(G,H);
 526             Heps             = _mm_mul_pd(vfeps,H);
 527             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 528             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 529             fvdw12           = _mm_mul_pd(c12_00,FF);
 530             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 531
 532             fscal            = fvdw;
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm_mul_pd(fscal,dx00);
 536             ty               = _mm_mul_pd(fscal,dy00);
 537             tz               = _mm_mul_pd(fscal,dz00);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_add_pd(fix0,tx);
 541             fiy0             = _mm_add_pd(fiy0,ty);
 542             fiz0             = _mm_add_pd(fiz0,tz);
 543
 544             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 545
 546             /* Inner loop uses 48 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             jnrA             = jjnr[jidx];
 553             j_coord_offsetA  = DIM*jnrA;
 554
 555             /* load j atom coordinates */
 556             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm_sub_pd(ix0,jx0);
 561             dy00             = _mm_sub_pd(iy0,jy0);
 562             dz00             = _mm_sub_pd(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 566
 567             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 568
 569             /* Load parameters for j particles */
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 580
 581             /* Calculate table index by multiplying r with table scale and truncate to integer */
 582             rt               = _mm_mul_pd(r00,vftabscale);
 583             vfitab           = _mm_cvttpd_epi32(rt);
 584             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 585             vfitab           = _mm_slli_epi32(vfitab,3);
 586
 587             /* CUBIC SPLINE TABLE DISPERSION */
 588             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 589             F                = _mm_setzero_pd();
 590             GMX_MM_TRANSPOSE2_PD(Y,F);
 591             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 592             H                = _mm_setzero_pd();
 593             GMX_MM_TRANSPOSE2_PD(G,H);
 594             Heps             = _mm_mul_pd(vfeps,H);
 595             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 596             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 597             fvdw6            = _mm_mul_pd(c6_00,FF);
 598
 599             /* CUBIC SPLINE TABLE REPULSION */
 600             vfitab           = _mm_add_epi32(vfitab,ifour);
 601             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 602             F                = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(Y,F);
 604             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 605             H                = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(G,H);
 607             Heps             = _mm_mul_pd(vfeps,H);
 608             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 609             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 610             fvdw12           = _mm_mul_pd(c12_00,FF);
 611             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 612
 613             fscal            = fvdw;
 614
 615             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_pd(fscal,dx00);
 619             ty               = _mm_mul_pd(fscal,dy00);
 620             tz               = _mm_mul_pd(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_add_pd(fix0,tx);
 624             fiy0             = _mm_add_pd(fiy0,ty);
 625             fiz0             = _mm_add_pd(fiz0,tz);
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 628
 629             /* Inner loop uses 48 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 6 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 649 }