src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_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_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_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     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  80     real             rswitch_scalar,d_scalar;
  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     facel            = _mm_set1_pd(fr->epsfac);
  97     charge           = mdatoms->chargeA;
  98     krf              = _mm_set1_pd(fr->ic->k_rf);
  99     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 100     crf              = _mm_set1_pd(fr->ic->c_rf);
 101     nvdwtype         = fr->ntype;
 102     vdwparam         = fr->nbfp;
 103     vdwtype          = mdatoms->typeA;
 104
 105     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 106     rcutoff_scalar   = fr->rcoulomb;
 107     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 108     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 109
 110     rswitch_scalar   = fr->rvdw_switch;
 111     rswitch          = _mm_set1_pd(rswitch_scalar);
 112     /* Setup switch parameters */
 113     d_scalar         = rcutoff_scalar-rswitch_scalar;
 114     d                = _mm_set1_pd(d_scalar);
 115     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 116     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 117     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 118     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 119     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 120     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_pd();
 148         fiy0             = _mm_setzero_pd();
 149         fiz0             = _mm_setzero_pd();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157         vvdwsum          = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 182
 183             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             if (gmx_mm_any_lt(rsq00,rcutoff2))
 195             {
 196
 197             r00              = _mm_mul_pd(rsq00,rinv00);
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm_mul_pd(iq0,jq0);
 201             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 203
 204             /* REACTION-FIELD ELECTROSTATICS */
 205             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 206             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 207
 208             /* LENNARD-JONES DISPERSION/REPULSION */
 209
 210             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 211             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 212             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 213             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 214             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 215
 216             d                = _mm_sub_pd(r00,rswitch);
 217             d                = _mm_max_pd(d,_mm_setzero_pd());
 218             d2               = _mm_mul_pd(d,d);
 219             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)))))));
 220
 221             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 222
 223             /* Evaluate switch function */
 224             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 225             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 226             vvdw             = _mm_mul_pd(vvdw,sw);
 227             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velec            = _mm_and_pd(velec,cutoff_mask);
 231             velecsum         = _mm_add_pd(velecsum,velec);
 232             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 233             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = _mm_add_pd(felec,fvdw);
 236
 237             fscal            = _mm_and_pd(fscal,cutoff_mask);
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm_mul_pd(fscal,dx00);
 241             ty               = _mm_mul_pd(fscal,dy00);
 242             tz               = _mm_mul_pd(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_add_pd(fix0,tx);
 246             fiy0             = _mm_add_pd(fiy0,ty);
 247             fiz0             = _mm_add_pd(fiz0,tz);
 248
 249             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 250
 251             }
 252
 253             /* Inner loop uses 70 flops */
 254         }
 255
 256         if(jidx<j_index_end)
 257         {
 258
 259             jnrA             = jjnr[jidx];
 260             j_coord_offsetA  = DIM*jnrA;
 261
 262             /* load j atom coordinates */
 263             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 264                                               &jx0,&jy0,&jz0);
 265
 266             /* Calculate displacement vector */
 267             dx00             = _mm_sub_pd(ix0,jx0);
 268             dy00             = _mm_sub_pd(iy0,jy0);
 269             dz00             = _mm_sub_pd(iz0,jz0);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 273
 274             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 275
 276             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 277
 278             /* Load parameters for j particles */
 279             jq0              = _mm_load_sd(charge+jnrA+0);
 280             vdwjidx0A        = 2*vdwtype[jnrA+0];
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             if (gmx_mm_any_lt(rsq00,rcutoff2))
 287             {
 288
 289             r00              = _mm_mul_pd(rsq00,rinv00);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq00             = _mm_mul_pd(iq0,jq0);
 293             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 294
 295             /* REACTION-FIELD ELECTROSTATICS */
 296             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 297             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 298
 299             /* LENNARD-JONES DISPERSION/REPULSION */
 300
 301             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 302             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 303             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 304             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 305             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 306
 307             d                = _mm_sub_pd(r00,rswitch);
 308             d                = _mm_max_pd(d,_mm_setzero_pd());
 309             d2               = _mm_mul_pd(d,d);
 310             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)))))));
 311
 312             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 313
 314             /* Evaluate switch function */
 315             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 316             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 317             vvdw             = _mm_mul_pd(vvdw,sw);
 318             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velec            = _mm_and_pd(velec,cutoff_mask);
 322             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 325             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 326             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 327
 328             fscal            = _mm_add_pd(felec,fvdw);
 329
 330             fscal            = _mm_and_pd(fscal,cutoff_mask);
 331
 332             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm_mul_pd(fscal,dx00);
 336             ty               = _mm_mul_pd(fscal,dy00);
 337             tz               = _mm_mul_pd(fscal,dz00);
 338
 339             /* Update vectorial force */
 340             fix0             = _mm_add_pd(fix0,tx);
 341             fiy0             = _mm_add_pd(fiy0,ty);
 342             fiz0             = _mm_add_pd(fiz0,tz);
 343
 344             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 345
 346             }
 347
 348             /* Inner loop uses 70 flops */
 349         }
 350
 351         /* End of innermost loop */
 352
 353         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 354                                               f+i_coord_offset,fshift+i_shift_offset);
 355
 356         ggid                        = gid[iidx];
 357         /* Update potential energies */
 358         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 359         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 360
 361         /* Increment number of inner iterations */
 362         inneriter                  += j_index_end - j_index_start;
 363
 364         /* Outer loop uses 9 flops */
 365     }
 366
 367     /* Increment number of outer iterations */
 368     outeriter        += nri;
 369
 370     /* Update outer/inner flops */
 371
 372     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 373 }
 374 /*
 375  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_double
 376  * Electrostatics interaction: ReactionField
 377  * VdW interaction:            LennardJones
 378  * Geometry:                   Particle-Particle
 379  * Calculate force/pot:        Force
 380  */
 381 void
 382 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_double
 383                     (t_nblist * gmx_restrict                nlist,
 384                      rvec * gmx_restrict                    xx,
 385                      rvec * gmx_restrict                    ff,
 386                      t_forcerec * gmx_restrict              fr,
 387                      t_mdatoms * gmx_restrict               mdatoms,
 388                      nb_kernel_data_t * gmx_restrict        kernel_data,
 389                      t_nrnb * gmx_restrict                  nrnb)
 390 {
 391     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 392      * just 0 for non-waters.
 393      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 394      * jnr indices corresponding to data put in the four positions in the SIMD register.
 395      */
 396     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 397     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 398     int              jnrA,jnrB;
 399     int              j_coord_offsetA,j_coord_offsetB;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 404     int              vdwioffset0;
 405     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 406     int              vdwjidx0A,vdwjidx0B;
 407     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 408     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 409     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 410     real             *charge;
 411     int              nvdwtype;
 412     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 413     int              *vdwtype;
 414     real             *vdwparam;
 415     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 416     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 417     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 418     real             rswitch_scalar,d_scalar;
 419     __m128d          dummy_mask,cutoff_mask;
 420     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 421     __m128d          one     = _mm_set1_pd(1.0);
 422     __m128d          two     = _mm_set1_pd(2.0);
 423     x                = xx[0];
 424     f                = ff[0];
 425
 426     nri              = nlist->nri;
 427     iinr             = nlist->iinr;
 428     jindex           = nlist->jindex;
 429     jjnr             = nlist->jjnr;
 430     shiftidx         = nlist->shift;
 431     gid              = nlist->gid;
 432     shiftvec         = fr->shift_vec[0];
 433     fshift           = fr->fshift[0];
 434     facel            = _mm_set1_pd(fr->epsfac);
 435     charge           = mdatoms->chargeA;
 436     krf              = _mm_set1_pd(fr->ic->k_rf);
 437     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 438     crf              = _mm_set1_pd(fr->ic->c_rf);
 439     nvdwtype         = fr->ntype;
 440     vdwparam         = fr->nbfp;
 441     vdwtype          = mdatoms->typeA;
 442
 443     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 444     rcutoff_scalar   = fr->rcoulomb;
 445     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 446     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 447
 448     rswitch_scalar   = fr->rvdw_switch;
 449     rswitch          = _mm_set1_pd(rswitch_scalar);
 450     /* Setup switch parameters */
 451     d_scalar         = rcutoff_scalar-rswitch_scalar;
 452     d                = _mm_set1_pd(d_scalar);
 453     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 454     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 455     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 456     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 457     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 458     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 459
 460     /* Avoid stupid compiler warnings */
 461     jnrA = jnrB = 0;
 462     j_coord_offsetA = 0;
 463     j_coord_offsetB = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     /* Start outer loop over neighborlists */
 469     for(iidx=0; iidx<nri; iidx++)
 470     {
 471         /* Load shift vector for this list */
 472         i_shift_offset   = DIM*shiftidx[iidx];
 473
 474         /* Load limits for loop over neighbors */
 475         j_index_start    = jindex[iidx];
 476         j_index_end      = jindex[iidx+1];
 477
 478         /* Get outer coordinate index */
 479         inr              = iinr[iidx];
 480         i_coord_offset   = DIM*inr;
 481
 482         /* Load i particle coords and add shift vector */
 483         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 484
 485         fix0             = _mm_setzero_pd();
 486         fiy0             = _mm_setzero_pd();
 487         fiz0             = _mm_setzero_pd();
 488
 489         /* Load parameters for i particles */
 490         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 491         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502
 503             /* load j atom coordinates */
 504             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                               &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm_sub_pd(ix0,jx0);
 509             dy00             = _mm_sub_pd(iy0,jy0);
 510             dz00             = _mm_sub_pd(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 516
 517             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             if (gmx_mm_any_lt(rsq00,rcutoff2))
 529             {
 530
 531             r00              = _mm_mul_pd(rsq00,rinv00);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq00             = _mm_mul_pd(iq0,jq0);
 535             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 536                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 540
 541             /* LENNARD-JONES DISPERSION/REPULSION */
 542
 543             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 544             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 545             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 546             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 547             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 548
 549             d                = _mm_sub_pd(r00,rswitch);
 550             d                = _mm_max_pd(d,_mm_setzero_pd());
 551             d2               = _mm_mul_pd(d,d);
 552             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)))))));
 553
 554             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 555
 556             /* Evaluate switch function */
 557             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 558             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 559             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 560
 561             fscal            = _mm_add_pd(felec,fvdw);
 562
 563             fscal            = _mm_and_pd(fscal,cutoff_mask);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_pd(fscal,dx00);
 567             ty               = _mm_mul_pd(fscal,dy00);
 568             tz               = _mm_mul_pd(fscal,dz00);
 569
 570             /* Update vectorial force */
 571             fix0             = _mm_add_pd(fix0,tx);
 572             fiy0             = _mm_add_pd(fiy0,ty);
 573             fiz0             = _mm_add_pd(fiz0,tz);
 574
 575             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 576
 577             }
 578
 579             /* Inner loop uses 61 flops */
 580         }
 581
 582         if(jidx<j_index_end)
 583         {
 584
 585             jnrA             = jjnr[jidx];
 586             j_coord_offsetA  = DIM*jnrA;
 587
 588             /* load j atom coordinates */
 589             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 590                                               &jx0,&jy0,&jz0);
 591
 592             /* Calculate displacement vector */
 593             dx00             = _mm_sub_pd(ix0,jx0);
 594             dy00             = _mm_sub_pd(iy0,jy0);
 595             dz00             = _mm_sub_pd(iz0,jz0);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 599
 600             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 601
 602             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 603
 604             /* Load parameters for j particles */
 605             jq0              = _mm_load_sd(charge+jnrA+0);
 606             vdwjidx0A        = 2*vdwtype[jnrA+0];
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             if (gmx_mm_any_lt(rsq00,rcutoff2))
 613             {
 614
 615             r00              = _mm_mul_pd(rsq00,rinv00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 620
 621             /* REACTION-FIELD ELECTROSTATICS */
 622             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 623
 624             /* LENNARD-JONES DISPERSION/REPULSION */
 625
 626             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 627             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 628             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 629             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 630             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 631
 632             d                = _mm_sub_pd(r00,rswitch);
 633             d                = _mm_max_pd(d,_mm_setzero_pd());
 634             d2               = _mm_mul_pd(d,d);
 635             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)))))));
 636
 637             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 638
 639             /* Evaluate switch function */
 640             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 641             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 642             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 643
 644             fscal            = _mm_add_pd(felec,fvdw);
 645
 646             fscal            = _mm_and_pd(fscal,cutoff_mask);
 647
 648             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm_mul_pd(fscal,dx00);
 652             ty               = _mm_mul_pd(fscal,dy00);
 653             tz               = _mm_mul_pd(fscal,dz00);
 654
 655             /* Update vectorial force */
 656             fix0             = _mm_add_pd(fix0,tx);
 657             fiy0             = _mm_add_pd(fiy0,ty);
 658             fiz0             = _mm_add_pd(fiz0,tz);
 659
 660             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 661
 662             }
 663
 664             /* Inner loop uses 61 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         /* Increment number of inner iterations */
 673         inneriter                  += j_index_end - j_index_start;
 674
 675         /* Outer loop uses 7 flops */
 676     }
 677
 678     /* Increment number of outer iterations */
 679     outeriter        += nri;
 680
 681     /* Update outer/inner flops */
 682
 683     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 684 }