src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm_set1_ps(fr->ic->k_rf);
 118     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 119     crf              = _mm_set1_ps(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->rvdw_switch;
 130     rswitch          = _mm_set1_ps(rswitch_scalar);
 131     /* Setup switch parameters */
 132     d_scalar         = rcutoff_scalar-rswitch_scalar;
 133     d                = _mm_set1_ps(d_scalar);
 134     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 135     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 138     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176
 177         /* Load parameters for i particles */
 178         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 179         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211
 212             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 213
 214             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 218                                                               charge+jnrC+0,charge+jnrD+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221             vdwjidx0C        = 2*vdwtype[jnrC+0];
 222             vdwjidx0D        = 2*vdwtype[jnrD+0];
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             if (gmx_mm_any_lt(rsq00,rcutoff2))
 229             {
 230
 231             r00              = _mm_mul_ps(rsq00,rinv00);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq00             = _mm_mul_ps(iq0,jq0);
 235             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 236                                          vdwparam+vdwioffset0+vdwjidx0B,
 237                                          vdwparam+vdwioffset0+vdwjidx0C,
 238                                          vdwparam+vdwioffset0+vdwjidx0D,
 239                                          &c6_00,&c12_00);
 240
 241             /* REACTION-FIELD ELECTROSTATICS */
 242             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 243             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 249             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 250             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 251             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 252
 253             d                = _mm_sub_ps(r00,rswitch);
 254             d                = _mm_max_ps(d,_mm_setzero_ps());
 255             d2               = _mm_mul_ps(d,d);
 256             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)))))));
 257
 258             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 259
 260             /* Evaluate switch function */
 261             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 262             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 263             vvdw             = _mm_mul_ps(vvdw,sw);
 264             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velec            = _mm_and_ps(velec,cutoff_mask);
 268             velecsum         = _mm_add_ps(velecsum,velec);
 269             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 270             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_ps(felec,fvdw);
 273
 274             fscal            = _mm_and_ps(fscal,cutoff_mask);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_ps(fscal,dx00);
 278             ty               = _mm_mul_ps(fscal,dy00);
 279             tz               = _mm_mul_ps(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_add_ps(fix0,tx);
 283             fiy0             = _mm_add_ps(fiy0,ty);
 284             fiz0             = _mm_add_ps(fiz0,tz);
 285
 286             fjptrA             = f+j_coord_offsetA;
 287             fjptrB             = f+j_coord_offsetB;
 288             fjptrC             = f+j_coord_offsetC;
 289             fjptrD             = f+j_coord_offsetD;
 290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 291
 292             }
 293
 294             /* Inner loop uses 70 flops */
 295         }
 296
 297         if(jidx<j_index_end)
 298         {
 299
 300             /* Get j neighbor index, and coordinate index */
 301             jnrlistA         = jjnr[jidx];
 302             jnrlistB         = jjnr[jidx+1];
 303             jnrlistC         = jjnr[jidx+2];
 304             jnrlistD         = jjnr[jidx+3];
 305             /* Sign of each element will be negative for non-real atoms.
 306              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 307              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 308              */
 309             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 310             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 311             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 312             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 313             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 314             j_coord_offsetA  = DIM*jnrA;
 315             j_coord_offsetB  = DIM*jnrB;
 316             j_coord_offsetC  = DIM*jnrC;
 317             j_coord_offsetD  = DIM*jnrD;
 318
 319             /* load j atom coordinates */
 320             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 321                                               x+j_coord_offsetC,x+j_coord_offsetD,
 322                                               &jx0,&jy0,&jz0);
 323
 324             /* Calculate displacement vector */
 325             dx00             = _mm_sub_ps(ix0,jx0);
 326             dy00             = _mm_sub_ps(iy0,jy0);
 327             dz00             = _mm_sub_ps(iz0,jz0);
 328
 329             /* Calculate squared distance and things based on it */
 330             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 331
 332             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 333
 334             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 335
 336             /* Load parameters for j particles */
 337             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 338                                                               charge+jnrC+0,charge+jnrD+0);
 339             vdwjidx0A        = 2*vdwtype[jnrA+0];
 340             vdwjidx0B        = 2*vdwtype[jnrB+0];
 341             vdwjidx0C        = 2*vdwtype[jnrC+0];
 342             vdwjidx0D        = 2*vdwtype[jnrD+0];
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             if (gmx_mm_any_lt(rsq00,rcutoff2))
 349             {
 350
 351             r00              = _mm_mul_ps(rsq00,rinv00);
 352             r00              = _mm_andnot_ps(dummy_mask,r00);
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq00             = _mm_mul_ps(iq0,jq0);
 356             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 357                                          vdwparam+vdwioffset0+vdwjidx0B,
 358                                          vdwparam+vdwioffset0+vdwjidx0C,
 359                                          vdwparam+vdwioffset0+vdwjidx0D,
 360                                          &c6_00,&c12_00);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 364             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 365
 366             /* LENNARD-JONES DISPERSION/REPULSION */
 367
 368             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 369             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 370             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 371             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 372             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 373
 374             d                = _mm_sub_ps(r00,rswitch);
 375             d                = _mm_max_ps(d,_mm_setzero_ps());
 376             d2               = _mm_mul_ps(d,d);
 377             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)))))));
 378
 379             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 380
 381             /* Evaluate switch function */
 382             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 383             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 384             vvdw             = _mm_mul_ps(vvdw,sw);
 385             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_and_ps(velec,cutoff_mask);
 389             velec            = _mm_andnot_ps(dummy_mask,velec);
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 392             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 393             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 394
 395             fscal            = _mm_add_ps(felec,fvdw);
 396
 397             fscal            = _mm_and_ps(fscal,cutoff_mask);
 398
 399             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_ps(fscal,dx00);
 403             ty               = _mm_mul_ps(fscal,dy00);
 404             tz               = _mm_mul_ps(fscal,dz00);
 405
 406             /* Update vectorial force */
 407             fix0             = _mm_add_ps(fix0,tx);
 408             fiy0             = _mm_add_ps(fiy0,ty);
 409             fiz0             = _mm_add_ps(fiz0,tz);
 410
 411             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 412             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 413             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 414             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 416
 417             }
 418
 419             /* Inner loop uses 71 flops */
 420         }
 421
 422         /* End of innermost loop */
 423
 424         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 425                                               f+i_coord_offset,fshift+i_shift_offset);
 426
 427         ggid                        = gid[iidx];
 428         /* Update potential energies */
 429         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 430         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 431
 432         /* Increment number of inner iterations */
 433         inneriter                  += j_index_end - j_index_start;
 434
 435         /* Outer loop uses 9 flops */
 436     }
 437
 438     /* Increment number of outer iterations */
 439     outeriter        += nri;
 440
 441     /* Update outer/inner flops */
 442
 443     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*71);
 444 }
 445 /*
 446  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_single
 447  * Electrostatics interaction: ReactionField
 448  * VdW interaction:            LennardJones
 449  * Geometry:                   Particle-Particle
 450  * Calculate force/pot:        Force
 451  */
 452 void
 453 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_single
 454                     (t_nblist                    * gmx_restrict       nlist,
 455                      rvec                        * gmx_restrict          xx,
 456                      rvec                        * gmx_restrict          ff,
 457                      t_forcerec                  * gmx_restrict          fr,
 458                      t_mdatoms                   * gmx_restrict     mdatoms,
 459                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 460                      t_nrnb                      * gmx_restrict        nrnb)
 461 {
 462     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 463      * just 0 for non-waters.
 464      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 465      * jnr indices corresponding to data put in the four positions in the SIMD register.
 466      */
 467     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 468     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 469     int              jnrA,jnrB,jnrC,jnrD;
 470     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 471     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 472     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 473     real             rcutoff_scalar;
 474     real             *shiftvec,*fshift,*x,*f;
 475     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 476     real             scratch[4*DIM];
 477     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 478     int              vdwioffset0;
 479     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 480     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 481     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 482     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 483     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 484     real             *charge;
 485     int              nvdwtype;
 486     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 487     int              *vdwtype;
 488     real             *vdwparam;
 489     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 490     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 491     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 492     real             rswitch_scalar,d_scalar;
 493     __m128           dummy_mask,cutoff_mask;
 494     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 495     __m128           one     = _mm_set1_ps(1.0);
 496     __m128           two     = _mm_set1_ps(2.0);
 497     x                = xx[0];
 498     f                = ff[0];
 499
 500     nri              = nlist->nri;
 501     iinr             = nlist->iinr;
 502     jindex           = nlist->jindex;
 503     jjnr             = nlist->jjnr;
 504     shiftidx         = nlist->shift;
 505     gid              = nlist->gid;
 506     shiftvec         = fr->shift_vec[0];
 507     fshift           = fr->fshift[0];
 508     facel            = _mm_set1_ps(fr->epsfac);
 509     charge           = mdatoms->chargeA;
 510     krf              = _mm_set1_ps(fr->ic->k_rf);
 511     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 512     crf              = _mm_set1_ps(fr->ic->c_rf);
 513     nvdwtype         = fr->ntype;
 514     vdwparam         = fr->nbfp;
 515     vdwtype          = mdatoms->typeA;
 516
 517     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 518     rcutoff_scalar   = fr->rcoulomb;
 519     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 520     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 521
 522     rswitch_scalar   = fr->rvdw_switch;
 523     rswitch          = _mm_set1_ps(rswitch_scalar);
 524     /* Setup switch parameters */
 525     d_scalar         = rcutoff_scalar-rswitch_scalar;
 526     d                = _mm_set1_ps(d_scalar);
 527     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 528     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 529     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 530     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 531     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 532     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 533
 534     /* Avoid stupid compiler warnings */
 535     jnrA = jnrB = jnrC = jnrD = 0;
 536     j_coord_offsetA = 0;
 537     j_coord_offsetB = 0;
 538     j_coord_offsetC = 0;
 539     j_coord_offsetD = 0;
 540
 541     outeriter        = 0;
 542     inneriter        = 0;
 543
 544     for(iidx=0;iidx<4*DIM;iidx++)
 545     {
 546         scratch[iidx] = 0.0;
 547     }
 548
 549     /* Start outer loop over neighborlists */
 550     for(iidx=0; iidx<nri; iidx++)
 551     {
 552         /* Load shift vector for this list */
 553         i_shift_offset   = DIM*shiftidx[iidx];
 554
 555         /* Load limits for loop over neighbors */
 556         j_index_start    = jindex[iidx];
 557         j_index_end      = jindex[iidx+1];
 558
 559         /* Get outer coordinate index */
 560         inr              = iinr[iidx];
 561         i_coord_offset   = DIM*inr;
 562
 563         /* Load i particle coords and add shift vector */
 564         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 565
 566         fix0             = _mm_setzero_ps();
 567         fiy0             = _mm_setzero_ps();
 568         fiz0             = _mm_setzero_ps();
 569
 570         /* Load parameters for i particles */
 571         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 572         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 573
 574         /* Start inner kernel loop */
 575         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 576         {
 577
 578             /* Get j neighbor index, and coordinate index */
 579             jnrA             = jjnr[jidx];
 580             jnrB             = jjnr[jidx+1];
 581             jnrC             = jjnr[jidx+2];
 582             jnrD             = jjnr[jidx+3];
 583             j_coord_offsetA  = DIM*jnrA;
 584             j_coord_offsetB  = DIM*jnrB;
 585             j_coord_offsetC  = DIM*jnrC;
 586             j_coord_offsetD  = DIM*jnrD;
 587
 588             /* load j atom coordinates */
 589             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 590                                               x+j_coord_offsetC,x+j_coord_offsetD,
 591                                               &jx0,&jy0,&jz0);
 592
 593             /* Calculate displacement vector */
 594             dx00             = _mm_sub_ps(ix0,jx0);
 595             dy00             = _mm_sub_ps(iy0,jy0);
 596             dz00             = _mm_sub_ps(iz0,jz0);
 597
 598             /* Calculate squared distance and things based on it */
 599             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 600
 601             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 602
 603             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 604
 605             /* Load parameters for j particles */
 606             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 607                                                               charge+jnrC+0,charge+jnrD+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609             vdwjidx0B        = 2*vdwtype[jnrB+0];
 610             vdwjidx0C        = 2*vdwtype[jnrC+0];
 611             vdwjidx0D        = 2*vdwtype[jnrD+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq00,rcutoff2))
 618             {
 619
 620             r00              = _mm_mul_ps(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_ps(iq0,jq0);
 624             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 625                                          vdwparam+vdwioffset0+vdwjidx0B,
 626                                          vdwparam+vdwioffset0+vdwjidx0C,
 627                                          vdwparam+vdwioffset0+vdwjidx0D,
 628                                          &c6_00,&c12_00);
 629
 630             /* REACTION-FIELD ELECTROSTATICS */
 631             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 632
 633             /* LENNARD-JONES DISPERSION/REPULSION */
 634
 635             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 636             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 637             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 638             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 639             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 640
 641             d                = _mm_sub_ps(r00,rswitch);
 642             d                = _mm_max_ps(d,_mm_setzero_ps());
 643             d2               = _mm_mul_ps(d,d);
 644             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)))))));
 645
 646             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 647
 648             /* Evaluate switch function */
 649             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 650             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 651             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 652
 653             fscal            = _mm_add_ps(felec,fvdw);
 654
 655             fscal            = _mm_and_ps(fscal,cutoff_mask);
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm_mul_ps(fscal,dx00);
 659             ty               = _mm_mul_ps(fscal,dy00);
 660             tz               = _mm_mul_ps(fscal,dz00);
 661
 662             /* Update vectorial force */
 663             fix0             = _mm_add_ps(fix0,tx);
 664             fiy0             = _mm_add_ps(fiy0,ty);
 665             fiz0             = _mm_add_ps(fiz0,tz);
 666
 667             fjptrA             = f+j_coord_offsetA;
 668             fjptrB             = f+j_coord_offsetB;
 669             fjptrC             = f+j_coord_offsetC;
 670             fjptrD             = f+j_coord_offsetD;
 671             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 672
 673             }
 674
 675             /* Inner loop uses 61 flops */
 676         }
 677
 678         if(jidx<j_index_end)
 679         {
 680
 681             /* Get j neighbor index, and coordinate index */
 682             jnrlistA         = jjnr[jidx];
 683             jnrlistB         = jjnr[jidx+1];
 684             jnrlistC         = jjnr[jidx+2];
 685             jnrlistD         = jjnr[jidx+3];
 686             /* Sign of each element will be negative for non-real atoms.
 687              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 688              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 689              */
 690             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 691             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 692             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 693             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 694             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 695             j_coord_offsetA  = DIM*jnrA;
 696             j_coord_offsetB  = DIM*jnrB;
 697             j_coord_offsetC  = DIM*jnrC;
 698             j_coord_offsetD  = DIM*jnrD;
 699
 700             /* load j atom coordinates */
 701             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 702                                               x+j_coord_offsetC,x+j_coord_offsetD,
 703                                               &jx0,&jy0,&jz0);
 704
 705             /* Calculate displacement vector */
 706             dx00             = _mm_sub_ps(ix0,jx0);
 707             dy00             = _mm_sub_ps(iy0,jy0);
 708             dz00             = _mm_sub_ps(iz0,jz0);
 709
 710             /* Calculate squared distance and things based on it */
 711             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 712
 713             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 714
 715             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 716
 717             /* Load parameters for j particles */
 718             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 719                                                               charge+jnrC+0,charge+jnrD+0);
 720             vdwjidx0A        = 2*vdwtype[jnrA+0];
 721             vdwjidx0B        = 2*vdwtype[jnrB+0];
 722             vdwjidx0C        = 2*vdwtype[jnrC+0];
 723             vdwjidx0D        = 2*vdwtype[jnrD+0];
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             if (gmx_mm_any_lt(rsq00,rcutoff2))
 730             {
 731
 732             r00              = _mm_mul_ps(rsq00,rinv00);
 733             r00              = _mm_andnot_ps(dummy_mask,r00);
 734
 735             /* Compute parameters for interactions between i and j atoms */
 736             qq00             = _mm_mul_ps(iq0,jq0);
 737             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 738                                          vdwparam+vdwioffset0+vdwjidx0B,
 739                                          vdwparam+vdwioffset0+vdwjidx0C,
 740                                          vdwparam+vdwioffset0+vdwjidx0D,
 741                                          &c6_00,&c12_00);
 742
 743             /* REACTION-FIELD ELECTROSTATICS */
 744             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 745
 746             /* LENNARD-JONES DISPERSION/REPULSION */
 747
 748             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 749             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 750             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 751             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 752             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 753
 754             d                = _mm_sub_ps(r00,rswitch);
 755             d                = _mm_max_ps(d,_mm_setzero_ps());
 756             d2               = _mm_mul_ps(d,d);
 757             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)))))));
 758
 759             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 760
 761             /* Evaluate switch function */
 762             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 763             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 764             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 765
 766             fscal            = _mm_add_ps(felec,fvdw);
 767
 768             fscal            = _mm_and_ps(fscal,cutoff_mask);
 769
 770             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 771
 772             /* Calculate temporary vectorial force */
 773             tx               = _mm_mul_ps(fscal,dx00);
 774             ty               = _mm_mul_ps(fscal,dy00);
 775             tz               = _mm_mul_ps(fscal,dz00);
 776
 777             /* Update vectorial force */
 778             fix0             = _mm_add_ps(fix0,tx);
 779             fiy0             = _mm_add_ps(fiy0,ty);
 780             fiz0             = _mm_add_ps(fiz0,tz);
 781
 782             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 783             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 784             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 785             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 786             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 787
 788             }
 789
 790             /* Inner loop uses 62 flops */
 791         }
 792
 793         /* End of innermost loop */
 794
 795         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 796                                               f+i_coord_offset,fshift+i_shift_offset);
 797
 798         /* Increment number of inner iterations */
 799         inneriter                  += j_index_end - j_index_start;
 800
 801         /* Outer loop uses 7 flops */
 802     }
 803
 804     /* Increment number of outer iterations */
 805     outeriter        += nri;
 806
 807     /* Update outer/inner flops */
 808
 809     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 810 }