src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_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     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  78     real             rswitch_scalar,d_scalar;
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     nvdwtype         = fr->ntype;
  95     vdwparam         = fr->nbfp;
  96     vdwtype          = mdatoms->typeA;
  97
  98     rcutoff_scalar   = fr->rvdw;
  99     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 100     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 101
 102     rswitch_scalar   = fr->rvdw_switch;
 103     rswitch          = _mm_set1_pd(rswitch_scalar);
 104     /* Setup switch parameters */
 105     d_scalar         = rcutoff_scalar-rswitch_scalar;
 106     d                = _mm_set1_pd(d_scalar);
 107     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 108     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 109     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 110     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 111     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 112     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     /* Start outer loop over neighborlists */
 123     for(iidx=0; iidx<nri; iidx++)
 124     {
 125         /* Load shift vector for this list */
 126         i_shift_offset   = DIM*shiftidx[iidx];
 127
 128         /* Load limits for loop over neighbors */
 129         j_index_start    = jindex[iidx];
 130         j_index_end      = jindex[iidx+1];
 131
 132         /* Get outer coordinate index */
 133         inr              = iinr[iidx];
 134         i_coord_offset   = DIM*inr;
 135
 136         /* Load i particle coords and add shift vector */
 137         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 138
 139         fix0             = _mm_setzero_pd();
 140         fiy0             = _mm_setzero_pd();
 141         fiz0             = _mm_setzero_pd();
 142
 143         /* Load parameters for i particles */
 144         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146         /* Reset potential sums */
 147         vvdwsum          = _mm_setzero_pd();
 148
 149         /* Start inner kernel loop */
 150         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 151         {
 152
 153             /* Get j neighbor index, and coordinate index */
 154             jnrA             = jjnr[jidx];
 155             jnrB             = jjnr[jidx+1];
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158
 159             /* load j atom coordinates */
 160             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 161                                               &jx0,&jy0,&jz0);
 162
 163             /* Calculate displacement vector */
 164             dx00             = _mm_sub_pd(ix0,jx0);
 165             dy00             = _mm_sub_pd(iy0,jy0);
 166             dz00             = _mm_sub_pd(iz0,jz0);
 167
 168             /* Calculate squared distance and things based on it */
 169             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 170
 171             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 172
 173             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 193
 194             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 195             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 196             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 197             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 198             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 199
 200             d                = _mm_sub_pd(r00,rswitch);
 201             d                = _mm_max_pd(d,_mm_setzero_pd());
 202             d2               = _mm_mul_pd(d,d);
 203             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 204
 205             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 206
 207             /* Evaluate switch function */
 208             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 209             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 210             vvdw             = _mm_mul_pd(vvdw,sw);
 211             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 215             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 216
 217             fscal            = fvdw;
 218
 219             fscal            = _mm_and_pd(fscal,cutoff_mask);
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm_mul_pd(fscal,dx00);
 223             ty               = _mm_mul_pd(fscal,dy00);
 224             tz               = _mm_mul_pd(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm_add_pd(fix0,tx);
 228             fiy0             = _mm_add_pd(fiy0,ty);
 229             fiz0             = _mm_add_pd(fiz0,tz);
 230
 231             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 232
 233             }
 234
 235             /* Inner loop uses 59 flops */
 236         }
 237
 238         if(jidx<j_index_end)
 239         {
 240
 241             jnrA             = jjnr[jidx];
 242             j_coord_offsetA  = DIM*jnrA;
 243
 244             /* load j atom coordinates */
 245             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 246                                               &jx0,&jy0,&jz0);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm_sub_pd(ix0,jx0);
 250             dy00             = _mm_sub_pd(iy0,jy0);
 251             dz00             = _mm_sub_pd(iz0,jz0);
 252
 253             /* Calculate squared distance and things based on it */
 254             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 255
 256             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 257
 258             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 259
 260             /* Load parameters for j particles */
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             r00              = _mm_mul_pd(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 279             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 280             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 281             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 282
 283             d                = _mm_sub_pd(r00,rswitch);
 284             d                = _mm_max_pd(d,_mm_setzero_pd());
 285             d2               = _mm_mul_pd(d,d);
 286             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 287
 288             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 289
 290             /* Evaluate switch function */
 291             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 292             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 293             vvdw             = _mm_mul_pd(vvdw,sw);
 294             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 298             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 299             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 300
 301             fscal            = fvdw;
 302
 303             fscal            = _mm_and_pd(fscal,cutoff_mask);
 304
 305             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_pd(fscal,dx00);
 309             ty               = _mm_mul_pd(fscal,dy00);
 310             tz               = _mm_mul_pd(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_add_pd(fix0,tx);
 314             fiy0             = _mm_add_pd(fiy0,ty);
 315             fiz0             = _mm_add_pd(fiz0,tz);
 316
 317             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 318
 319             }
 320
 321             /* Inner loop uses 59 flops */
 322         }
 323
 324         /* End of innermost loop */
 325
 326         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 327                                               f+i_coord_offset,fshift+i_shift_offset);
 328
 329         ggid                        = gid[iidx];
 330         /* Update potential energies */
 331         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 332
 333         /* Increment number of inner iterations */
 334         inneriter                  += j_index_end - j_index_start;
 335
 336         /* Outer loop uses 7 flops */
 337     }
 338
 339     /* Increment number of outer iterations */
 340     outeriter        += nri;
 341
 342     /* Update outer/inner flops */
 343
 344     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 345 }
 346 /*
 347  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_double
 348  * Electrostatics interaction: None
 349  * VdW interaction:            LennardJones
 350  * Geometry:                   Particle-Particle
 351  * Calculate force/pot:        Force
 352  */
 353 void
 354 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_double
 355                     (t_nblist * gmx_restrict                nlist,
 356                      rvec * gmx_restrict                    xx,
 357                      rvec * gmx_restrict                    ff,
 358                      t_forcerec * gmx_restrict              fr,
 359                      t_mdatoms * gmx_restrict               mdatoms,
 360                      nb_kernel_data_t * gmx_restrict        kernel_data,
 361                      t_nrnb * gmx_restrict                  nrnb)
 362 {
 363     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 364      * just 0 for non-waters.
 365      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 366      * jnr indices corresponding to data put in the four positions in the SIMD register.
 367      */
 368     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 369     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 370     int              jnrA,jnrB;
 371     int              j_coord_offsetA,j_coord_offsetB;
 372     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 373     real             rcutoff_scalar;
 374     real             *shiftvec,*fshift,*x,*f;
 375     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 376     int              vdwioffset0;
 377     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 378     int              vdwjidx0A,vdwjidx0B;
 379     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 380     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 381     int              nvdwtype;
 382     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 383     int              *vdwtype;
 384     real             *vdwparam;
 385     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 386     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 387     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 388     real             rswitch_scalar,d_scalar;
 389     __m128d          dummy_mask,cutoff_mask;
 390     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 391     __m128d          one     = _mm_set1_pd(1.0);
 392     __m128d          two     = _mm_set1_pd(2.0);
 393     x                = xx[0];
 394     f                = ff[0];
 395
 396     nri              = nlist->nri;
 397     iinr             = nlist->iinr;
 398     jindex           = nlist->jindex;
 399     jjnr             = nlist->jjnr;
 400     shiftidx         = nlist->shift;
 401     gid              = nlist->gid;
 402     shiftvec         = fr->shift_vec[0];
 403     fshift           = fr->fshift[0];
 404     nvdwtype         = fr->ntype;
 405     vdwparam         = fr->nbfp;
 406     vdwtype          = mdatoms->typeA;
 407
 408     rcutoff_scalar   = fr->rvdw;
 409     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 410     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 411
 412     rswitch_scalar   = fr->rvdw_switch;
 413     rswitch          = _mm_set1_pd(rswitch_scalar);
 414     /* Setup switch parameters */
 415     d_scalar         = rcutoff_scalar-rswitch_scalar;
 416     d                = _mm_set1_pd(d_scalar);
 417     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 418     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 419     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 420     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 421     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 422     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 423
 424     /* Avoid stupid compiler warnings */
 425     jnrA = jnrB = 0;
 426     j_coord_offsetA = 0;
 427     j_coord_offsetB = 0;
 428
 429     outeriter        = 0;
 430     inneriter        = 0;
 431
 432     /* Start outer loop over neighborlists */
 433     for(iidx=0; iidx<nri; iidx++)
 434     {
 435         /* Load shift vector for this list */
 436         i_shift_offset   = DIM*shiftidx[iidx];
 437
 438         /* Load limits for loop over neighbors */
 439         j_index_start    = jindex[iidx];
 440         j_index_end      = jindex[iidx+1];
 441
 442         /* Get outer coordinate index */
 443         inr              = iinr[iidx];
 444         i_coord_offset   = DIM*inr;
 445
 446         /* Load i particle coords and add shift vector */
 447         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 448
 449         fix0             = _mm_setzero_pd();
 450         fiy0             = _mm_setzero_pd();
 451         fiz0             = _mm_setzero_pd();
 452
 453         /* Load parameters for i particles */
 454         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 455
 456         /* Start inner kernel loop */
 457         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 458         {
 459
 460             /* Get j neighbor index, and coordinate index */
 461             jnrA             = jjnr[jidx];
 462             jnrB             = jjnr[jidx+1];
 463             j_coord_offsetA  = DIM*jnrA;
 464             j_coord_offsetB  = DIM*jnrB;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                               &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm_sub_pd(ix0,jx0);
 472             dy00             = _mm_sub_pd(iy0,jy0);
 473             dz00             = _mm_sub_pd(iz0,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 477
 478             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 479
 480             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 481
 482             /* Load parameters for j particles */
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484             vdwjidx0B        = 2*vdwtype[jnrB+0];
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             if (gmx_mm_any_lt(rsq00,rcutoff2))
 491             {
 492
 493             r00              = _mm_mul_pd(rsq00,rinv00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 497                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 498
 499             /* LENNARD-JONES DISPERSION/REPULSION */
 500
 501             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 502             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 503             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 504             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 505             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 506
 507             d                = _mm_sub_pd(r00,rswitch);
 508             d                = _mm_max_pd(d,_mm_setzero_pd());
 509             d2               = _mm_mul_pd(d,d);
 510             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 511
 512             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 513
 514             /* Evaluate switch function */
 515             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 516             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 517             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 518
 519             fscal            = fvdw;
 520
 521             fscal            = _mm_and_pd(fscal,cutoff_mask);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm_mul_pd(fscal,dx00);
 525             ty               = _mm_mul_pd(fscal,dy00);
 526             tz               = _mm_mul_pd(fscal,dz00);
 527
 528             /* Update vectorial force */
 529             fix0             = _mm_add_pd(fix0,tx);
 530             fiy0             = _mm_add_pd(fiy0,ty);
 531             fiz0             = _mm_add_pd(fiz0,tz);
 532
 533             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 534
 535             }
 536
 537             /* Inner loop uses 56 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             jnrA             = jjnr[jidx];
 544             j_coord_offsetA  = DIM*jnrA;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 548                                               &jx0,&jy0,&jz0);
 549
 550             /* Calculate displacement vector */
 551             dx00             = _mm_sub_pd(ix0,jx0);
 552             dy00             = _mm_sub_pd(iy0,jy0);
 553             dz00             = _mm_sub_pd(iz0,jz0);
 554
 555             /* Calculate squared distance and things based on it */
 556             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 557
 558             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 559
 560             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 561
 562             /* Load parameters for j particles */
 563             vdwjidx0A        = 2*vdwtype[jnrA+0];
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq00,rcutoff2))
 570             {
 571
 572             r00              = _mm_mul_pd(rsq00,rinv00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 580             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 581             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 582             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 583             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 584
 585             d                = _mm_sub_pd(r00,rswitch);
 586             d                = _mm_max_pd(d,_mm_setzero_pd());
 587             d2               = _mm_mul_pd(d,d);
 588             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 589
 590             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 591
 592             /* Evaluate switch function */
 593             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 594             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 595             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 596
 597             fscal            = fvdw;
 598
 599             fscal            = _mm_and_pd(fscal,cutoff_mask);
 600
 601             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 602
 603             /* Calculate temporary vectorial force */
 604             tx               = _mm_mul_pd(fscal,dx00);
 605             ty               = _mm_mul_pd(fscal,dy00);
 606             tz               = _mm_mul_pd(fscal,dz00);
 607
 608             /* Update vectorial force */
 609             fix0             = _mm_add_pd(fix0,tx);
 610             fiy0             = _mm_add_pd(fiy0,ty);
 611             fiz0             = _mm_add_pd(fiz0,tz);
 612
 613             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 614
 615             }
 616
 617             /* Inner loop uses 56 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 6 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*56);
 637 }