src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     rswitch_scalar   = fr->rvdw_switch;
 128     rswitch          = _mm_set1_ps(rswitch_scalar);
 129     /* Setup switch parameters */
 130     d_scalar         = rcutoff_scalar-rswitch_scalar;
 131     d                = _mm_set1_ps(d_scalar);
 132     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 133     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 134     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 135     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 136     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174
 175         /* Load parameters for i particles */
 176         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 177         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 209
 210             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 211
 212             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             if (gmx_mm_any_lt(rsq00,rcutoff2))
 227             {
 228
 229             r00              = _mm_mul_ps(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_ps(iq0,jq0);
 233             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,
 235                                          vdwparam+vdwioffset0+vdwjidx0C,
 236                                          vdwparam+vdwioffset0+vdwjidx0D,
 237                                          &c6_00,&c12_00);
 238
 239             /* REACTION-FIELD ELECTROSTATICS */
 240             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 241             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 248             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 250
 251             d                = _mm_sub_ps(r00,rswitch);
 252             d                = _mm_max_ps(d,_mm_setzero_ps());
 253             d2               = _mm_mul_ps(d,d);
 254             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)))))));
 255
 256             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 257
 258             /* Evaluate switch function */
 259             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 260             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 261             vvdw             = _mm_mul_ps(vvdw,sw);
 262             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velec            = _mm_and_ps(velec,cutoff_mask);
 266             velecsum         = _mm_add_ps(velecsum,velec);
 267             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 268             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 269
 270             fscal            = _mm_add_ps(felec,fvdw);
 271
 272             fscal            = _mm_and_ps(fscal,cutoff_mask);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_ps(fscal,dx00);
 276             ty               = _mm_mul_ps(fscal,dy00);
 277             tz               = _mm_mul_ps(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_ps(fix0,tx);
 281             fiy0             = _mm_add_ps(fiy0,ty);
 282             fiz0             = _mm_add_ps(fiz0,tz);
 283
 284             fjptrA             = f+j_coord_offsetA;
 285             fjptrB             = f+j_coord_offsetB;
 286             fjptrC             = f+j_coord_offsetC;
 287             fjptrD             = f+j_coord_offsetD;
 288             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 289
 290             }
 291
 292             /* Inner loop uses 70 flops */
 293         }
 294
 295         if(jidx<j_index_end)
 296         {
 297
 298             /* Get j neighbor index, and coordinate index */
 299             jnrlistA         = jjnr[jidx];
 300             jnrlistB         = jjnr[jidx+1];
 301             jnrlistC         = jjnr[jidx+2];
 302             jnrlistD         = jjnr[jidx+3];
 303             /* Sign of each element will be negative for non-real atoms.
 304              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 305              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 306              */
 307             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 308             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 309             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 310             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 311             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 312             j_coord_offsetA  = DIM*jnrA;
 313             j_coord_offsetB  = DIM*jnrB;
 314             j_coord_offsetC  = DIM*jnrC;
 315             j_coord_offsetD  = DIM*jnrD;
 316
 317             /* load j atom coordinates */
 318             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 319                                               x+j_coord_offsetC,x+j_coord_offsetD,
 320                                               &jx0,&jy0,&jz0);
 321
 322             /* Calculate displacement vector */
 323             dx00             = _mm_sub_ps(ix0,jx0);
 324             dy00             = _mm_sub_ps(iy0,jy0);
 325             dz00             = _mm_sub_ps(iz0,jz0);
 326
 327             /* Calculate squared distance and things based on it */
 328             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 329
 330             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 331
 332             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 333
 334             /* Load parameters for j particles */
 335             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 336                                                               charge+jnrC+0,charge+jnrD+0);
 337             vdwjidx0A        = 2*vdwtype[jnrA+0];
 338             vdwjidx0B        = 2*vdwtype[jnrB+0];
 339             vdwjidx0C        = 2*vdwtype[jnrC+0];
 340             vdwjidx0D        = 2*vdwtype[jnrD+0];
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm_any_lt(rsq00,rcutoff2))
 347             {
 348
 349             r00              = _mm_mul_ps(rsq00,rinv00);
 350             r00              = _mm_andnot_ps(dummy_mask,r00);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq00             = _mm_mul_ps(iq0,jq0);
 354             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 355                                          vdwparam+vdwioffset0+vdwjidx0B,
 356                                          vdwparam+vdwioffset0+vdwjidx0C,
 357                                          vdwparam+vdwioffset0+vdwjidx0D,
 358                                          &c6_00,&c12_00);
 359
 360             /* REACTION-FIELD ELECTROSTATICS */
 361             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 362             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 363
 364             /* LENNARD-JONES DISPERSION/REPULSION */
 365
 366             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 367             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 368             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 369             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 370             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 371
 372             d                = _mm_sub_ps(r00,rswitch);
 373             d                = _mm_max_ps(d,_mm_setzero_ps());
 374             d2               = _mm_mul_ps(d,d);
 375             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)))))));
 376
 377             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 378
 379             /* Evaluate switch function */
 380             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 381             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 382             vvdw             = _mm_mul_ps(vvdw,sw);
 383             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm_and_ps(velec,cutoff_mask);
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 390             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 391             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 392
 393             fscal            = _mm_add_ps(felec,fvdw);
 394
 395             fscal            = _mm_and_ps(fscal,cutoff_mask);
 396
 397             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx00);
 401             ty               = _mm_mul_ps(fscal,dy00);
 402             tz               = _mm_mul_ps(fscal,dz00);
 403
 404             /* Update vectorial force */
 405             fix0             = _mm_add_ps(fix0,tx);
 406             fiy0             = _mm_add_ps(fiy0,ty);
 407             fiz0             = _mm_add_ps(fiz0,tz);
 408
 409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 413             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 414
 415             }
 416
 417             /* Inner loop uses 71 flops */
 418         }
 419
 420         /* End of innermost loop */
 421
 422         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 423                                               f+i_coord_offset,fshift+i_shift_offset);
 424
 425         ggid                        = gid[iidx];
 426         /* Update potential energies */
 427         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 428         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 429
 430         /* Increment number of inner iterations */
 431         inneriter                  += j_index_end - j_index_start;
 432
 433         /* Outer loop uses 9 flops */
 434     }
 435
 436     /* Increment number of outer iterations */
 437     outeriter        += nri;
 438
 439     /* Update outer/inner flops */
 440
 441     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*71);
 442 }
 443 /*
 444  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_single
 445  * Electrostatics interaction: ReactionField
 446  * VdW interaction:            LennardJones
 447  * Geometry:                   Particle-Particle
 448  * Calculate force/pot:        Force
 449  */
 450 void
 451 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_single
 452                     (t_nblist                    * gmx_restrict       nlist,
 453                      rvec                        * gmx_restrict          xx,
 454                      rvec                        * gmx_restrict          ff,
 455                      t_forcerec                  * gmx_restrict          fr,
 456                      t_mdatoms                   * gmx_restrict     mdatoms,
 457                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 458                      t_nrnb                      * gmx_restrict        nrnb)
 459 {
 460     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 461      * just 0 for non-waters.
 462      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 463      * jnr indices corresponding to data put in the four positions in the SIMD register.
 464      */
 465     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 466     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 467     int              jnrA,jnrB,jnrC,jnrD;
 468     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 469     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 470     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 471     real             rcutoff_scalar;
 472     real             *shiftvec,*fshift,*x,*f;
 473     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 474     real             scratch[4*DIM];
 475     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 476     int              vdwioffset0;
 477     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 478     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 479     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 480     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 481     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 482     real             *charge;
 483     int              nvdwtype;
 484     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 485     int              *vdwtype;
 486     real             *vdwparam;
 487     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 488     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 489     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 490     real             rswitch_scalar,d_scalar;
 491     __m128           dummy_mask,cutoff_mask;
 492     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 493     __m128           one     = _mm_set1_ps(1.0);
 494     __m128           two     = _mm_set1_ps(2.0);
 495     x                = xx[0];
 496     f                = ff[0];
 497
 498     nri              = nlist->nri;
 499     iinr             = nlist->iinr;
 500     jindex           = nlist->jindex;
 501     jjnr             = nlist->jjnr;
 502     shiftidx         = nlist->shift;
 503     gid              = nlist->gid;
 504     shiftvec         = fr->shift_vec[0];
 505     fshift           = fr->fshift[0];
 506     facel            = _mm_set1_ps(fr->epsfac);
 507     charge           = mdatoms->chargeA;
 508     krf              = _mm_set1_ps(fr->ic->k_rf);
 509     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 510     crf              = _mm_set1_ps(fr->ic->c_rf);
 511     nvdwtype         = fr->ntype;
 512     vdwparam         = fr->nbfp;
 513     vdwtype          = mdatoms->typeA;
 514
 515     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 516     rcutoff_scalar   = fr->rcoulomb;
 517     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 518     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 519
 520     rswitch_scalar   = fr->rvdw_switch;
 521     rswitch          = _mm_set1_ps(rswitch_scalar);
 522     /* Setup switch parameters */
 523     d_scalar         = rcutoff_scalar-rswitch_scalar;
 524     d                = _mm_set1_ps(d_scalar);
 525     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 526     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 527     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 528     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 529     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 530     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 531
 532     /* Avoid stupid compiler warnings */
 533     jnrA = jnrB = jnrC = jnrD = 0;
 534     j_coord_offsetA = 0;
 535     j_coord_offsetB = 0;
 536     j_coord_offsetC = 0;
 537     j_coord_offsetD = 0;
 538
 539     outeriter        = 0;
 540     inneriter        = 0;
 541
 542     for(iidx=0;iidx<4*DIM;iidx++)
 543     {
 544         scratch[iidx] = 0.0;
 545     }
 546
 547     /* Start outer loop over neighborlists */
 548     for(iidx=0; iidx<nri; iidx++)
 549     {
 550         /* Load shift vector for this list */
 551         i_shift_offset   = DIM*shiftidx[iidx];
 552
 553         /* Load limits for loop over neighbors */
 554         j_index_start    = jindex[iidx];
 555         j_index_end      = jindex[iidx+1];
 556
 557         /* Get outer coordinate index */
 558         inr              = iinr[iidx];
 559         i_coord_offset   = DIM*inr;
 560
 561         /* Load i particle coords and add shift vector */
 562         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 563
 564         fix0             = _mm_setzero_ps();
 565         fiy0             = _mm_setzero_ps();
 566         fiz0             = _mm_setzero_ps();
 567
 568         /* Load parameters for i particles */
 569         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 570         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 571
 572         /* Start inner kernel loop */
 573         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 574         {
 575
 576             /* Get j neighbor index, and coordinate index */
 577             jnrA             = jjnr[jidx];
 578             jnrB             = jjnr[jidx+1];
 579             jnrC             = jjnr[jidx+2];
 580             jnrD             = jjnr[jidx+3];
 581             j_coord_offsetA  = DIM*jnrA;
 582             j_coord_offsetB  = DIM*jnrB;
 583             j_coord_offsetC  = DIM*jnrC;
 584             j_coord_offsetD  = DIM*jnrD;
 585
 586             /* load j atom coordinates */
 587             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 588                                               x+j_coord_offsetC,x+j_coord_offsetD,
 589                                               &jx0,&jy0,&jz0);
 590
 591             /* Calculate displacement vector */
 592             dx00             = _mm_sub_ps(ix0,jx0);
 593             dy00             = _mm_sub_ps(iy0,jy0);
 594             dz00             = _mm_sub_ps(iz0,jz0);
 595
 596             /* Calculate squared distance and things based on it */
 597             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 598
 599             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 600
 601             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 602
 603             /* Load parameters for j particles */
 604             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 605                                                               charge+jnrC+0,charge+jnrD+0);
 606             vdwjidx0A        = 2*vdwtype[jnrA+0];
 607             vdwjidx0B        = 2*vdwtype[jnrB+0];
 608             vdwjidx0C        = 2*vdwtype[jnrC+0];
 609             vdwjidx0D        = 2*vdwtype[jnrD+0];
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm_any_lt(rsq00,rcutoff2))
 616             {
 617
 618             r00              = _mm_mul_ps(rsq00,rinv00);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm_mul_ps(iq0,jq0);
 622             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 623                                          vdwparam+vdwioffset0+vdwjidx0B,
 624                                          vdwparam+vdwioffset0+vdwjidx0C,
 625                                          vdwparam+vdwioffset0+vdwjidx0D,
 626                                          &c6_00,&c12_00);
 627
 628             /* REACTION-FIELD ELECTROSTATICS */
 629             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 630
 631             /* LENNARD-JONES DISPERSION/REPULSION */
 632
 633             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 634             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 635             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 636             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 637             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 638
 639             d                = _mm_sub_ps(r00,rswitch);
 640             d                = _mm_max_ps(d,_mm_setzero_ps());
 641             d2               = _mm_mul_ps(d,d);
 642             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)))))));
 643
 644             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 645
 646             /* Evaluate switch function */
 647             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 648             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 649             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 650
 651             fscal            = _mm_add_ps(felec,fvdw);
 652
 653             fscal            = _mm_and_ps(fscal,cutoff_mask);
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm_mul_ps(fscal,dx00);
 657             ty               = _mm_mul_ps(fscal,dy00);
 658             tz               = _mm_mul_ps(fscal,dz00);
 659
 660             /* Update vectorial force */
 661             fix0             = _mm_add_ps(fix0,tx);
 662             fiy0             = _mm_add_ps(fiy0,ty);
 663             fiz0             = _mm_add_ps(fiz0,tz);
 664
 665             fjptrA             = f+j_coord_offsetA;
 666             fjptrB             = f+j_coord_offsetB;
 667             fjptrC             = f+j_coord_offsetC;
 668             fjptrD             = f+j_coord_offsetD;
 669             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 670
 671             }
 672
 673             /* Inner loop uses 61 flops */
 674         }
 675
 676         if(jidx<j_index_end)
 677         {
 678
 679             /* Get j neighbor index, and coordinate index */
 680             jnrlistA         = jjnr[jidx];
 681             jnrlistB         = jjnr[jidx+1];
 682             jnrlistC         = jjnr[jidx+2];
 683             jnrlistD         = jjnr[jidx+3];
 684             /* Sign of each element will be negative for non-real atoms.
 685              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 686              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 687              */
 688             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 689             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 690             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 691             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 692             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 693             j_coord_offsetA  = DIM*jnrA;
 694             j_coord_offsetB  = DIM*jnrB;
 695             j_coord_offsetC  = DIM*jnrC;
 696             j_coord_offsetD  = DIM*jnrD;
 697
 698             /* load j atom coordinates */
 699             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 700                                               x+j_coord_offsetC,x+j_coord_offsetD,
 701                                               &jx0,&jy0,&jz0);
 702
 703             /* Calculate displacement vector */
 704             dx00             = _mm_sub_ps(ix0,jx0);
 705             dy00             = _mm_sub_ps(iy0,jy0);
 706             dz00             = _mm_sub_ps(iz0,jz0);
 707
 708             /* Calculate squared distance and things based on it */
 709             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 710
 711             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 712
 713             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 714
 715             /* Load parameters for j particles */
 716             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 717                                                               charge+jnrC+0,charge+jnrD+0);
 718             vdwjidx0A        = 2*vdwtype[jnrA+0];
 719             vdwjidx0B        = 2*vdwtype[jnrB+0];
 720             vdwjidx0C        = 2*vdwtype[jnrC+0];
 721             vdwjidx0D        = 2*vdwtype[jnrD+0];
 722
 723             /**************************
 724              * CALCULATE INTERACTIONS *
 725              **************************/
 726
 727             if (gmx_mm_any_lt(rsq00,rcutoff2))
 728             {
 729
 730             r00              = _mm_mul_ps(rsq00,rinv00);
 731             r00              = _mm_andnot_ps(dummy_mask,r00);
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq00             = _mm_mul_ps(iq0,jq0);
 735             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 736                                          vdwparam+vdwioffset0+vdwjidx0B,
 737                                          vdwparam+vdwioffset0+vdwjidx0C,
 738                                          vdwparam+vdwioffset0+vdwjidx0D,
 739                                          &c6_00,&c12_00);
 740
 741             /* REACTION-FIELD ELECTROSTATICS */
 742             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 743
 744             /* LENNARD-JONES DISPERSION/REPULSION */
 745
 746             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 747             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 748             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 749             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 750             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 751
 752             d                = _mm_sub_ps(r00,rswitch);
 753             d                = _mm_max_ps(d,_mm_setzero_ps());
 754             d2               = _mm_mul_ps(d,d);
 755             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)))))));
 756
 757             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 758
 759             /* Evaluate switch function */
 760             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 761             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 762             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 763
 764             fscal            = _mm_add_ps(felec,fvdw);
 765
 766             fscal            = _mm_and_ps(fscal,cutoff_mask);
 767
 768             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 769
 770             /* Calculate temporary vectorial force */
 771             tx               = _mm_mul_ps(fscal,dx00);
 772             ty               = _mm_mul_ps(fscal,dy00);
 773             tz               = _mm_mul_ps(fscal,dz00);
 774
 775             /* Update vectorial force */
 776             fix0             = _mm_add_ps(fix0,tx);
 777             fiy0             = _mm_add_ps(fiy0,ty);
 778             fiz0             = _mm_add_ps(fiz0,tz);
 779
 780             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 781             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 782             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 783             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 784             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 785
 786             }
 787
 788             /* Inner loop uses 62 flops */
 789         }
 790
 791         /* End of innermost loop */
 792
 793         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 794                                               f+i_coord_offset,fshift+i_shift_offset);
 795
 796         /* Increment number of inner iterations */
 797         inneriter                  += j_index_end - j_index_start;
 798
 799         /* Outer loop uses 7 flops */
 800     }
 801
 802     /* Increment number of outer iterations */
 803     outeriter        += nri;
 804
 805     /* Update outer/inner flops */
 806
 807     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 808 }