src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single 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_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_single
  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_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     int              nvdwtype;
  72     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  76     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  77     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  78     real             rswitch_scalar,d_scalar;
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(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_ps(rcutoff_scalar);
 100     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 101
 102     rswitch_scalar   = fr->rvdw_switch;
 103     rswitch          = _mm_set1_ps(rswitch_scalar);
 104     /* Setup switch parameters */
 105     d_scalar         = rcutoff_scalar-rswitch_scalar;
 106     d                = _mm_set1_ps(d_scalar);
 107     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 108     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 109     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 110     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 111     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 112     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129         shX              = shiftvec[i_shift_offset+XX];
 130         shY              = shiftvec[i_shift_offset+YY];
 131         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 143         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 144         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 145
 146         fix0             = _mm_setzero_ps();
 147         fiy0             = _mm_setzero_ps();
 148         fiz0             = _mm_setzero_ps();
 149
 150         /* Load parameters for i particles */
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         vvdwsum          = _mm_setzero_ps();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             jnrC             = jjnr[jidx+2];
 164             jnrD             = jjnr[jidx+3];
 165
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 185
 186             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191             vdwjidx0C        = 2*vdwtype[jnrC+0];
 192             vdwjidx0D        = 2*vdwtype[jnrD+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             if (gmx_mm_any_lt(rsq00,rcutoff2))
 199             {
 200
 201             r00              = _mm_mul_ps(rsq00,rinv00);
 202
 203             /* Compute parameters for interactions between i and j atoms */
 204             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,
 206                                          vdwparam+vdwioffset0+vdwjidx0C,
 207                                          vdwparam+vdwioffset0+vdwjidx0D,
 208                                          &c6_00,&c12_00);
 209
 210             /* LENNARD-JONES DISPERSION/REPULSION */
 211
 212             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 213             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 214             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 215             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 216             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 217
 218             d                = _mm_sub_ps(r00,rswitch);
 219             d                = _mm_max_ps(d,_mm_setzero_ps());
 220             d2               = _mm_mul_ps(d,d);
 221             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 222
 223             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 224
 225             /* Evaluate switch function */
 226             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 227             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 228             vvdw             = _mm_mul_ps(vvdw,sw);
 229             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 233             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 234
 235             fscal            = fvdw;
 236
 237             fscal            = _mm_and_ps(fscal,cutoff_mask);
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm_mul_ps(fscal,dx00);
 241             ty               = _mm_mul_ps(fscal,dy00);
 242             tz               = _mm_mul_ps(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_add_ps(fix0,tx);
 246             fiy0             = _mm_add_ps(fiy0,ty);
 247             fiz0             = _mm_add_ps(fiz0,tz);
 248
 249             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 250                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 251                                                    tx,ty,tz);
 252
 253             }
 254
 255             /* Inner loop uses 59 flops */
 256         }
 257
 258         if(jidx<j_index_end)
 259         {
 260
 261             /* Get j neighbor index, and coordinate index */
 262             jnrA             = jjnr[jidx];
 263             jnrB             = jjnr[jidx+1];
 264             jnrC             = jjnr[jidx+2];
 265             jnrD             = jjnr[jidx+3];
 266
 267             /* Sign of each element will be negative for non-real atoms.
 268              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 269              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 270              */
 271             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 272             jnrA       = (jnrA>=0) ? jnrA : 0;
 273             jnrB       = (jnrB>=0) ? jnrB : 0;
 274             jnrC       = (jnrC>=0) ? jnrC : 0;
 275             jnrD       = (jnrD>=0) ? jnrD : 0;
 276
 277             j_coord_offsetA  = DIM*jnrA;
 278             j_coord_offsetB  = DIM*jnrB;
 279             j_coord_offsetC  = DIM*jnrC;
 280             j_coord_offsetD  = DIM*jnrD;
 281
 282             /* load j atom coordinates */
 283             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 284                                               x+j_coord_offsetC,x+j_coord_offsetD,
 285                                               &jx0,&jy0,&jz0);
 286
 287             /* Calculate displacement vector */
 288             dx00             = _mm_sub_ps(ix0,jx0);
 289             dy00             = _mm_sub_ps(iy0,jy0);
 290             dz00             = _mm_sub_ps(iz0,jz0);
 291
 292             /* Calculate squared distance and things based on it */
 293             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 294
 295             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 296
 297             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 298
 299             /* Load parameters for j particles */
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301             vdwjidx0B        = 2*vdwtype[jnrB+0];
 302             vdwjidx0C        = 2*vdwtype[jnrC+0];
 303             vdwjidx0D        = 2*vdwtype[jnrD+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq00,rcutoff2))
 310             {
 311
 312             r00              = _mm_mul_ps(rsq00,rinv00);
 313             r00              = _mm_andnot_ps(dummy_mask,r00);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 317                                          vdwparam+vdwioffset0+vdwjidx0B,
 318                                          vdwparam+vdwioffset0+vdwjidx0C,
 319                                          vdwparam+vdwioffset0+vdwjidx0D,
 320                                          &c6_00,&c12_00);
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 326             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 327             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 328             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 329
 330             d                = _mm_sub_ps(r00,rswitch);
 331             d                = _mm_max_ps(d,_mm_setzero_ps());
 332             d2               = _mm_mul_ps(d,d);
 333             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 334
 335             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 336
 337             /* Evaluate switch function */
 338             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 339             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 340             vvdw             = _mm_mul_ps(vvdw,sw);
 341             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 345             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 346             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 347
 348             fscal            = fvdw;
 349
 350             fscal            = _mm_and_ps(fscal,cutoff_mask);
 351
 352             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm_mul_ps(fscal,dx00);
 356             ty               = _mm_mul_ps(fscal,dy00);
 357             tz               = _mm_mul_ps(fscal,dz00);
 358
 359             /* Update vectorial force */
 360             fix0             = _mm_add_ps(fix0,tx);
 361             fiy0             = _mm_add_ps(fiy0,ty);
 362             fiz0             = _mm_add_ps(fiz0,tz);
 363
 364             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 365                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 366                                                    tx,ty,tz);
 367
 368             }
 369
 370             /* Inner loop uses 60 flops */
 371         }
 372
 373         /* End of innermost loop */
 374
 375         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 376                                               f+i_coord_offset,fshift+i_shift_offset);
 377
 378         ggid                        = gid[iidx];
 379         /* Update potential energies */
 380         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 381
 382         /* Increment number of inner iterations */
 383         inneriter                  += j_index_end - j_index_start;
 384
 385         /* Outer loop uses 10 flops */
 386     }
 387
 388     /* Increment number of outer iterations */
 389     outeriter        += nri;
 390
 391     /* Update outer/inner flops */
 392
 393     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*10 + inneriter*60);
 394 }
 395 /*
 396  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 397  * Electrostatics interaction: None
 398  * VdW interaction:            LennardJones
 399  * Geometry:                   Particle-Particle
 400  * Calculate force/pot:        Force
 401  */
 402 void
 403 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_single
 404                     (t_nblist * gmx_restrict                nlist,
 405                      rvec * gmx_restrict                    xx,
 406                      rvec * gmx_restrict                    ff,
 407                      t_forcerec * gmx_restrict              fr,
 408                      t_mdatoms * gmx_restrict               mdatoms,
 409                      nb_kernel_data_t * gmx_restrict        kernel_data,
 410                      t_nrnb * gmx_restrict                  nrnb)
 411 {
 412     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 413      * just 0 for non-waters.
 414      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 415      * jnr indices corresponding to data put in the four positions in the SIMD register.
 416      */
 417     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 418     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 419     int              jnrA,jnrB,jnrC,jnrD;
 420     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 421     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 422     real             shX,shY,shZ,rcutoff_scalar;
 423     real             *shiftvec,*fshift,*x,*f;
 424     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 425     int              vdwioffset0;
 426     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 427     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 428     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 429     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 430     int              nvdwtype;
 431     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 432     int              *vdwtype;
 433     real             *vdwparam;
 434     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 435     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 436     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 437     real             rswitch_scalar,d_scalar;
 438     __m128           dummy_mask,cutoff_mask;
 439     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 440     __m128           one     = _mm_set1_ps(1.0);
 441     __m128           two     = _mm_set1_ps(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456
 457     rcutoff_scalar   = fr->rvdw;
 458     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 459     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 460
 461     rswitch_scalar   = fr->rvdw_switch;
 462     rswitch          = _mm_set1_ps(rswitch_scalar);
 463     /* Setup switch parameters */
 464     d_scalar         = rcutoff_scalar-rswitch_scalar;
 465     d                = _mm_set1_ps(d_scalar);
 466     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 467     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 468     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 469     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 470     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 471     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 472
 473     /* Avoid stupid compiler warnings */
 474     jnrA = jnrB = jnrC = jnrD = 0;
 475     j_coord_offsetA = 0;
 476     j_coord_offsetB = 0;
 477     j_coord_offsetC = 0;
 478     j_coord_offsetD = 0;
 479
 480     outeriter        = 0;
 481     inneriter        = 0;
 482
 483     /* Start outer loop over neighborlists */
 484     for(iidx=0; iidx<nri; iidx++)
 485     {
 486         /* Load shift vector for this list */
 487         i_shift_offset   = DIM*shiftidx[iidx];
 488         shX              = shiftvec[i_shift_offset+XX];
 489         shY              = shiftvec[i_shift_offset+YY];
 490         shZ              = shiftvec[i_shift_offset+ZZ];
 491
 492         /* Load limits for loop over neighbors */
 493         j_index_start    = jindex[iidx];
 494         j_index_end      = jindex[iidx+1];
 495
 496         /* Get outer coordinate index */
 497         inr              = iinr[iidx];
 498         i_coord_offset   = DIM*inr;
 499
 500         /* Load i particle coords and add shift vector */
 501         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 502         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 503         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 504
 505         fix0             = _mm_setzero_ps();
 506         fiy0             = _mm_setzero_ps();
 507         fiz0             = _mm_setzero_ps();
 508
 509         /* Load parameters for i particles */
 510         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 511
 512         /* Start inner kernel loop */
 513         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 514         {
 515
 516             /* Get j neighbor index, and coordinate index */
 517             jnrA             = jjnr[jidx];
 518             jnrB             = jjnr[jidx+1];
 519             jnrC             = jjnr[jidx+2];
 520             jnrD             = jjnr[jidx+3];
 521
 522             j_coord_offsetA  = DIM*jnrA;
 523             j_coord_offsetB  = DIM*jnrB;
 524             j_coord_offsetC  = DIM*jnrC;
 525             j_coord_offsetD  = DIM*jnrD;
 526
 527             /* load j atom coordinates */
 528             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 529                                               x+j_coord_offsetC,x+j_coord_offsetD,
 530                                               &jx0,&jy0,&jz0);
 531
 532             /* Calculate displacement vector */
 533             dx00             = _mm_sub_ps(ix0,jx0);
 534             dy00             = _mm_sub_ps(iy0,jy0);
 535             dz00             = _mm_sub_ps(iz0,jz0);
 536
 537             /* Calculate squared distance and things based on it */
 538             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 539
 540             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 541
 542             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 543
 544             /* Load parameters for j particles */
 545             vdwjidx0A        = 2*vdwtype[jnrA+0];
 546             vdwjidx0B        = 2*vdwtype[jnrB+0];
 547             vdwjidx0C        = 2*vdwtype[jnrC+0];
 548             vdwjidx0D        = 2*vdwtype[jnrD+0];
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             if (gmx_mm_any_lt(rsq00,rcutoff2))
 555             {
 556
 557             r00              = _mm_mul_ps(rsq00,rinv00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,
 562                                          vdwparam+vdwioffset0+vdwjidx0C,
 563                                          vdwparam+vdwioffset0+vdwjidx0D,
 564                                          &c6_00,&c12_00);
 565
 566             /* LENNARD-JONES DISPERSION/REPULSION */
 567
 568             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 569             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 570             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 571             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 572             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 573
 574             d                = _mm_sub_ps(r00,rswitch);
 575             d                = _mm_max_ps(d,_mm_setzero_ps());
 576             d2               = _mm_mul_ps(d,d);
 577             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 578
 579             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 580
 581             /* Evaluate switch function */
 582             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 583             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 584             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 585
 586             fscal            = fvdw;
 587
 588             fscal            = _mm_and_ps(fscal,cutoff_mask);
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm_mul_ps(fscal,dx00);
 592             ty               = _mm_mul_ps(fscal,dy00);
 593             tz               = _mm_mul_ps(fscal,dz00);
 594
 595             /* Update vectorial force */
 596             fix0             = _mm_add_ps(fix0,tx);
 597             fiy0             = _mm_add_ps(fiy0,ty);
 598             fiz0             = _mm_add_ps(fiz0,tz);
 599
 600             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 601                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 602                                                    tx,ty,tz);
 603
 604             }
 605
 606             /* Inner loop uses 56 flops */
 607         }
 608
 609         if(jidx<j_index_end)
 610         {
 611
 612             /* Get j neighbor index, and coordinate index */
 613             jnrA             = jjnr[jidx];
 614             jnrB             = jjnr[jidx+1];
 615             jnrC             = jjnr[jidx+2];
 616             jnrD             = jjnr[jidx+3];
 617
 618             /* Sign of each element will be negative for non-real atoms.
 619              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 620              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 621              */
 622             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 623             jnrA       = (jnrA>=0) ? jnrA : 0;
 624             jnrB       = (jnrB>=0) ? jnrB : 0;
 625             jnrC       = (jnrC>=0) ? jnrC : 0;
 626             jnrD       = (jnrD>=0) ? jnrD : 0;
 627
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632
 633             /* load j atom coordinates */
 634             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 635                                               x+j_coord_offsetC,x+j_coord_offsetD,
 636                                               &jx0,&jy0,&jz0);
 637
 638             /* Calculate displacement vector */
 639             dx00             = _mm_sub_ps(ix0,jx0);
 640             dy00             = _mm_sub_ps(iy0,jy0);
 641             dz00             = _mm_sub_ps(iz0,jz0);
 642
 643             /* Calculate squared distance and things based on it */
 644             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 645
 646             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 647
 648             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 649
 650             /* Load parameters for j particles */
 651             vdwjidx0A        = 2*vdwtype[jnrA+0];
 652             vdwjidx0B        = 2*vdwtype[jnrB+0];
 653             vdwjidx0C        = 2*vdwtype[jnrC+0];
 654             vdwjidx0D        = 2*vdwtype[jnrD+0];
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm_any_lt(rsq00,rcutoff2))
 661             {
 662
 663             r00              = _mm_mul_ps(rsq00,rinv00);
 664             r00              = _mm_andnot_ps(dummy_mask,r00);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 668                                          vdwparam+vdwioffset0+vdwjidx0B,
 669                                          vdwparam+vdwioffset0+vdwjidx0C,
 670                                          vdwparam+vdwioffset0+vdwjidx0D,
 671                                          &c6_00,&c12_00);
 672
 673             /* LENNARD-JONES DISPERSION/REPULSION */
 674
 675             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 676             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 677             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 678             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 679             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 680
 681             d                = _mm_sub_ps(r00,rswitch);
 682             d                = _mm_max_ps(d,_mm_setzero_ps());
 683             d2               = _mm_mul_ps(d,d);
 684             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 685
 686             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 687
 688             /* Evaluate switch function */
 689             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 690             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 691             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 692
 693             fscal            = fvdw;
 694
 695             fscal            = _mm_and_ps(fscal,cutoff_mask);
 696
 697             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 698
 699             /* Calculate temporary vectorial force */
 700             tx               = _mm_mul_ps(fscal,dx00);
 701             ty               = _mm_mul_ps(fscal,dy00);
 702             tz               = _mm_mul_ps(fscal,dz00);
 703
 704             /* Update vectorial force */
 705             fix0             = _mm_add_ps(fix0,tx);
 706             fiy0             = _mm_add_ps(fiy0,ty);
 707             fiz0             = _mm_add_ps(fiz0,tz);
 708
 709             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 710                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 711                                                    tx,ty,tz);
 712
 713             }
 714
 715             /* Inner loop uses 57 flops */
 716         }
 717
 718         /* End of innermost loop */
 719
 720         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 721                                               f+i_coord_offset,fshift+i_shift_offset);
 722
 723         /* Increment number of inner iterations */
 724         inneriter                  += j_index_end - j_index_start;
 725
 726         /* Outer loop uses 9 flops */
 727     }
 728
 729     /* Increment number of outer iterations */
 730     outeriter        += nri;
 731
 732     /* Update outer/inner flops */
 733
 734     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*9 + inneriter*57);
 735 }