src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     int              nvdwtype;
  72     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  76     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  77     __m128i          vfitab;
  78     __m128i          ifour       = _mm_set1_epi32(4);
  79     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  80     real             *vftab;
  81     __m128d          dummy_mask,cutoff_mask;
  82     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  83     __m128d          one     = _mm_set1_pd(1.0);
  84     __m128d          two     = _mm_set1_pd(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     nvdwtype         = fr->ntype;
  97     vdwparam         = fr->nbfp;
  98     vdwtype          = mdatoms->typeA;
  99
 100     vftab            = kernel_data->table_vdw->data;
 101     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     /* Start outer loop over neighborlists */
 112     for(iidx=0; iidx<nri; iidx++)
 113     {
 114         /* Load shift vector for this list */
 115         i_shift_offset   = DIM*shiftidx[iidx];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 127
 128         fix0             = _mm_setzero_pd();
 129         fiy0             = _mm_setzero_pd();
 130         fiz0             = _mm_setzero_pd();
 131
 132         /* Load parameters for i particles */
 133         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135         /* Reset potential sums */
 136         vvdwsum          = _mm_setzero_pd();
 137
 138         /* Start inner kernel loop */
 139         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 140         {
 141
 142             /* Get j neighbor index, and coordinate index */
 143             jnrA             = jjnr[jidx];
 144             jnrB             = jjnr[jidx+1];
 145             j_coord_offsetA  = DIM*jnrA;
 146             j_coord_offsetB  = DIM*jnrB;
 147
 148             /* load j atom coordinates */
 149             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 150                                               &jx0,&jy0,&jz0);
 151
 152             /* Calculate displacement vector */
 153             dx00             = _mm_sub_pd(ix0,jx0);
 154             dy00             = _mm_sub_pd(iy0,jy0);
 155             dz00             = _mm_sub_pd(iz0,jz0);
 156
 157             /* Calculate squared distance and things based on it */
 158             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 159
 160             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 161
 162             /* Load parameters for j particles */
 163             vdwjidx0A        = 2*vdwtype[jnrA+0];
 164             vdwjidx0B        = 2*vdwtype[jnrB+0];
 165
 166             /**************************
 167              * CALCULATE INTERACTIONS *
 168              **************************/
 169
 170             r00              = _mm_mul_pd(rsq00,rinv00);
 171
 172             /* Compute parameters for interactions between i and j atoms */
 173             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 174                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 175
 176             /* Calculate table index by multiplying r with table scale and truncate to integer */
 177             rt               = _mm_mul_pd(r00,vftabscale);
 178             vfitab           = _mm_cvttpd_epi32(rt);
 179             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 180             vfitab           = _mm_slli_epi32(vfitab,3);
 181
 182             /* CUBIC SPLINE TABLE DISPERSION */
 183             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 184             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 185             GMX_MM_TRANSPOSE2_PD(Y,F);
 186             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 187             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 188             GMX_MM_TRANSPOSE2_PD(G,H);
 189             Heps             = _mm_mul_pd(vfeps,H);
 190             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 191             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 192             vvdw6            = _mm_mul_pd(c6_00,VV);
 193             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 194             fvdw6            = _mm_mul_pd(c6_00,FF);
 195
 196             /* CUBIC SPLINE TABLE REPULSION */
 197             vfitab           = _mm_add_epi32(vfitab,ifour);
 198             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 199             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 200             GMX_MM_TRANSPOSE2_PD(Y,F);
 201             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 202             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 203             GMX_MM_TRANSPOSE2_PD(G,H);
 204             Heps             = _mm_mul_pd(vfeps,H);
 205             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 206             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 207             vvdw12           = _mm_mul_pd(c12_00,VV);
 208             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 209             fvdw12           = _mm_mul_pd(c12_00,FF);
 210             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 211             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 215
 216             fscal            = fvdw;
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_pd(fscal,dx00);
 220             ty               = _mm_mul_pd(fscal,dy00);
 221             tz               = _mm_mul_pd(fscal,dz00);
 222
 223             /* Update vectorial force */
 224             fix0             = _mm_add_pd(fix0,tx);
 225             fiy0             = _mm_add_pd(fiy0,ty);
 226             fiz0             = _mm_add_pd(fiz0,tz);
 227
 228             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 229
 230             /* Inner loop uses 56 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             jnrA             = jjnr[jidx];
 237             j_coord_offsetA  = DIM*jnrA;
 238
 239             /* load j atom coordinates */
 240             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 241                                               &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm_sub_pd(ix0,jx0);
 245             dy00             = _mm_sub_pd(iy0,jy0);
 246             dz00             = _mm_sub_pd(iz0,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 250
 251             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 252
 253             /* Load parameters for j particles */
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             r00              = _mm_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 264
 265             /* Calculate table index by multiplying r with table scale and truncate to integer */
 266             rt               = _mm_mul_pd(r00,vftabscale);
 267             vfitab           = _mm_cvttpd_epi32(rt);
 268             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 269             vfitab           = _mm_slli_epi32(vfitab,3);
 270
 271             /* CUBIC SPLINE TABLE DISPERSION */
 272             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 273             F                = _mm_setzero_pd();
 274             GMX_MM_TRANSPOSE2_PD(Y,F);
 275             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 276             H                = _mm_setzero_pd();
 277             GMX_MM_TRANSPOSE2_PD(G,H);
 278             Heps             = _mm_mul_pd(vfeps,H);
 279             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 280             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 281             vvdw6            = _mm_mul_pd(c6_00,VV);
 282             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 283             fvdw6            = _mm_mul_pd(c6_00,FF);
 284
 285             /* CUBIC SPLINE TABLE REPULSION */
 286             vfitab           = _mm_add_epi32(vfitab,ifour);
 287             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 288             F                = _mm_setzero_pd();
 289             GMX_MM_TRANSPOSE2_PD(Y,F);
 290             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 291             H                = _mm_setzero_pd();
 292             GMX_MM_TRANSPOSE2_PD(G,H);
 293             Heps             = _mm_mul_pd(vfeps,H);
 294             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 295             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 296             vvdw12           = _mm_mul_pd(c12_00,VV);
 297             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 298             fvdw12           = _mm_mul_pd(c12_00,FF);
 299             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 300             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 304             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm_mul_pd(fscal,dx00);
 312             ty               = _mm_mul_pd(fscal,dy00);
 313             tz               = _mm_mul_pd(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_add_pd(fix0,tx);
 317             fiy0             = _mm_add_pd(fiy0,ty);
 318             fiz0             = _mm_add_pd(fiz0,tz);
 319
 320             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 321
 322             /* Inner loop uses 56 flops */
 323         }
 324
 325         /* End of innermost loop */
 326
 327         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 328                                               f+i_coord_offset,fshift+i_shift_offset);
 329
 330         ggid                        = gid[iidx];
 331         /* Update potential energies */
 332         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 333
 334         /* Increment number of inner iterations */
 335         inneriter                  += j_index_end - j_index_start;
 336
 337         /* Outer loop uses 7 flops */
 338     }
 339
 340     /* Increment number of outer iterations */
 341     outeriter        += nri;
 342
 343     /* Update outer/inner flops */
 344
 345     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*56);
 346 }
 347 /*
 348  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 349  * Electrostatics interaction: None
 350  * VdW interaction:            CubicSplineTable
 351  * Geometry:                   Particle-Particle
 352  * Calculate force/pot:        Force
 353  */
 354 void
 355 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_double
 356                     (t_nblist * gmx_restrict                nlist,
 357                      rvec * gmx_restrict                    xx,
 358                      rvec * gmx_restrict                    ff,
 359                      t_forcerec * gmx_restrict              fr,
 360                      t_mdatoms * gmx_restrict               mdatoms,
 361                      nb_kernel_data_t * gmx_restrict        kernel_data,
 362                      t_nrnb * gmx_restrict                  nrnb)
 363 {
 364     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 365      * just 0 for non-waters.
 366      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 367      * jnr indices corresponding to data put in the four positions in the SIMD register.
 368      */
 369     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 370     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 371     int              jnrA,jnrB;
 372     int              j_coord_offsetA,j_coord_offsetB;
 373     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 374     real             rcutoff_scalar;
 375     real             *shiftvec,*fshift,*x,*f;
 376     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 377     int              vdwioffset0;
 378     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 379     int              vdwjidx0A,vdwjidx0B;
 380     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 381     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 382     int              nvdwtype;
 383     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 384     int              *vdwtype;
 385     real             *vdwparam;
 386     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 387     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 388     __m128i          vfitab;
 389     __m128i          ifour       = _mm_set1_epi32(4);
 390     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 391     real             *vftab;
 392     __m128d          dummy_mask,cutoff_mask;
 393     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 394     __m128d          one     = _mm_set1_pd(1.0);
 395     __m128d          two     = _mm_set1_pd(2.0);
 396     x                = xx[0];
 397     f                = ff[0];
 398
 399     nri              = nlist->nri;
 400     iinr             = nlist->iinr;
 401     jindex           = nlist->jindex;
 402     jjnr             = nlist->jjnr;
 403     shiftidx         = nlist->shift;
 404     gid              = nlist->gid;
 405     shiftvec         = fr->shift_vec[0];
 406     fshift           = fr->fshift[0];
 407     nvdwtype         = fr->ntype;
 408     vdwparam         = fr->nbfp;
 409     vdwtype          = mdatoms->typeA;
 410
 411     vftab            = kernel_data->table_vdw->data;
 412     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 413
 414     /* Avoid stupid compiler warnings */
 415     jnrA = jnrB = 0;
 416     j_coord_offsetA = 0;
 417     j_coord_offsetB = 0;
 418
 419     outeriter        = 0;
 420     inneriter        = 0;
 421
 422     /* Start outer loop over neighborlists */
 423     for(iidx=0; iidx<nri; iidx++)
 424     {
 425         /* Load shift vector for this list */
 426         i_shift_offset   = DIM*shiftidx[iidx];
 427
 428         /* Load limits for loop over neighbors */
 429         j_index_start    = jindex[iidx];
 430         j_index_end      = jindex[iidx+1];
 431
 432         /* Get outer coordinate index */
 433         inr              = iinr[iidx];
 434         i_coord_offset   = DIM*inr;
 435
 436         /* Load i particle coords and add shift vector */
 437         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 438
 439         fix0             = _mm_setzero_pd();
 440         fiy0             = _mm_setzero_pd();
 441         fiz0             = _mm_setzero_pd();
 442
 443         /* Load parameters for i particles */
 444         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 445
 446         /* Start inner kernel loop */
 447         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 448         {
 449
 450             /* Get j neighbor index, and coordinate index */
 451             jnrA             = jjnr[jidx];
 452             jnrB             = jjnr[jidx+1];
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455
 456             /* load j atom coordinates */
 457             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_pd(ix0,jx0);
 462             dy00             = _mm_sub_pd(iy0,jy0);
 463             dz00             = _mm_sub_pd(iz0,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 467
 468             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 469
 470             /* Load parameters for j particles */
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472             vdwjidx0B        = 2*vdwtype[jnrB+0];
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             r00              = _mm_mul_pd(rsq00,rinv00);
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 482                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 483
 484             /* Calculate table index by multiplying r with table scale and truncate to integer */
 485             rt               = _mm_mul_pd(r00,vftabscale);
 486             vfitab           = _mm_cvttpd_epi32(rt);
 487             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 488             vfitab           = _mm_slli_epi32(vfitab,3);
 489
 490             /* CUBIC SPLINE TABLE DISPERSION */
 491             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 492             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 493             GMX_MM_TRANSPOSE2_PD(Y,F);
 494             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 495             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 496             GMX_MM_TRANSPOSE2_PD(G,H);
 497             Heps             = _mm_mul_pd(vfeps,H);
 498             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 499             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 500             fvdw6            = _mm_mul_pd(c6_00,FF);
 501
 502             /* CUBIC SPLINE TABLE REPULSION */
 503             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm_mul_pd(c12_00,FF);
 514             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 515
 516             fscal            = fvdw;
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx00);
 520             ty               = _mm_mul_pd(fscal,dy00);
 521             tz               = _mm_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_pd(fix0,tx);
 525             fiy0             = _mm_add_pd(fiy0,ty);
 526             fiz0             = _mm_add_pd(fiz0,tz);
 527
 528             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 529
 530             /* Inner loop uses 48 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             jnrA             = jjnr[jidx];
 537             j_coord_offsetA  = DIM*jnrA;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 541                                               &jx0,&jy0,&jz0);
 542
 543             /* Calculate displacement vector */
 544             dx00             = _mm_sub_pd(ix0,jx0);
 545             dy00             = _mm_sub_pd(iy0,jy0);
 546             dz00             = _mm_sub_pd(iz0,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 550
 551             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 552
 553             /* Load parameters for j particles */
 554             vdwjidx0A        = 2*vdwtype[jnrA+0];
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r00              = _mm_mul_pd(rsq00,rinv00);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 564
 565             /* Calculate table index by multiplying r with table scale and truncate to integer */
 566             rt               = _mm_mul_pd(r00,vftabscale);
 567             vfitab           = _mm_cvttpd_epi32(rt);
 568             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 569             vfitab           = _mm_slli_epi32(vfitab,3);
 570
 571             /* CUBIC SPLINE TABLE DISPERSION */
 572             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 573             F                = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(Y,F);
 575             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 576             H                = _mm_setzero_pd();
 577             GMX_MM_TRANSPOSE2_PD(G,H);
 578             Heps             = _mm_mul_pd(vfeps,H);
 579             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 580             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 581             fvdw6            = _mm_mul_pd(c6_00,FF);
 582
 583             /* CUBIC SPLINE TABLE REPULSION */
 584             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm_mul_pd(c12_00,FF);
 595             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 596
 597             fscal            = fvdw;
 598
 599             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_pd(fscal,dx00);
 603             ty               = _mm_mul_pd(fscal,dy00);
 604             tz               = _mm_mul_pd(fscal,dz00);
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_add_pd(fix0,tx);
 608             fiy0             = _mm_add_pd(fiy0,ty);
 609             fiz0             = _mm_add_pd(fiz0,tz);
 610
 611             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 612
 613             /* Inner loop uses 48 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 619                                               f+i_coord_offset,fshift+i_shift_offset);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 6 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 633 }