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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  96     real             rswitch_scalar,d_scalar;
  97     __m128           dummy_mask,cutoff_mask;
  98     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  99     __m128           one     = _mm_set1_ps(1.0);
 100     __m128           two     = _mm_set1_ps(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_ps(fr->ic->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_ps(fr->ic->k_rf);
 115     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_ps(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->ic->rcoulomb;
 123     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 125
 126     rswitch_scalar   = fr->ic->rvdw_switch;
 127     rswitch          = _mm_set1_ps(rswitch_scalar);
 128     /* Setup switch parameters */
 129     d_scalar         = rcutoff_scalar-rswitch_scalar;
 130     d                = _mm_set1_ps(d_scalar);
 131     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 132     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 135     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173
 174         /* Load parameters for i particles */
 175         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 176         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_ps();
 180         vvdwsum          = _mm_setzero_ps();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 208
 209             rinv00           = sse41_invsqrt_f(rsq00);
 210
 211             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 215                                                               charge+jnrC+0,charge+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             if (gmx_mm_any_lt(rsq00,rcutoff2))
 226             {
 227
 228             r00              = _mm_mul_ps(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm_mul_ps(iq0,jq0);
 232             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 233                                          vdwparam+vdwioffset0+vdwjidx0B,
 234                                          vdwparam+vdwioffset0+vdwjidx0C,
 235                                          vdwparam+vdwioffset0+vdwjidx0D,
 236                                          &c6_00,&c12_00);
 237
 238             /* REACTION-FIELD ELECTROSTATICS */
 239             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 240             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 241
 242             /* LENNARD-JONES DISPERSION/REPULSION */
 243
 244             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 245             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 246             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 247             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 248             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 249
 250             d                = _mm_sub_ps(r00,rswitch);
 251             d                = _mm_max_ps(d,_mm_setzero_ps());
 252             d2               = _mm_mul_ps(d,d);
 253             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)))))));
 254
 255             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 256
 257             /* Evaluate switch function */
 258             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 259             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 260             vvdw             = _mm_mul_ps(vvdw,sw);
 261             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velec            = _mm_and_ps(velec,cutoff_mask);
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 270
 271             fscal            = _mm_and_ps(fscal,cutoff_mask);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 288
 289             }
 290
 291             /* Inner loop uses 70 flops */
 292         }
 293
 294         if(jidx<j_index_end)
 295         {
 296
 297             /* Get j neighbor index, and coordinate index */
 298             jnrlistA         = jjnr[jidx];
 299             jnrlistB         = jjnr[jidx+1];
 300             jnrlistC         = jjnr[jidx+2];
 301             jnrlistD         = jjnr[jidx+3];
 302             /* Sign of each element will be negative for non-real atoms.
 303              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 304              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 305              */
 306             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 307             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 308             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 309             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 310             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 311             j_coord_offsetA  = DIM*jnrA;
 312             j_coord_offsetB  = DIM*jnrB;
 313             j_coord_offsetC  = DIM*jnrC;
 314             j_coord_offsetD  = DIM*jnrD;
 315
 316             /* load j atom coordinates */
 317             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 318                                               x+j_coord_offsetC,x+j_coord_offsetD,
 319                                               &jx0,&jy0,&jz0);
 320
 321             /* Calculate displacement vector */
 322             dx00             = _mm_sub_ps(ix0,jx0);
 323             dy00             = _mm_sub_ps(iy0,jy0);
 324             dz00             = _mm_sub_ps(iz0,jz0);
 325
 326             /* Calculate squared distance and things based on it */
 327             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 328
 329             rinv00           = sse41_invsqrt_f(rsq00);
 330
 331             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 332
 333             /* Load parameters for j particles */
 334             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 335                                                               charge+jnrC+0,charge+jnrD+0);
 336             vdwjidx0A        = 2*vdwtype[jnrA+0];
 337             vdwjidx0B        = 2*vdwtype[jnrB+0];
 338             vdwjidx0C        = 2*vdwtype[jnrC+0];
 339             vdwjidx0D        = 2*vdwtype[jnrD+0];
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq00,rcutoff2))
 346             {
 347
 348             r00              = _mm_mul_ps(rsq00,rinv00);
 349             r00              = _mm_andnot_ps(dummy_mask,r00);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq00             = _mm_mul_ps(iq0,jq0);
 353             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 354                                          vdwparam+vdwioffset0+vdwjidx0B,
 355                                          vdwparam+vdwioffset0+vdwjidx0C,
 356                                          vdwparam+vdwioffset0+vdwjidx0D,
 357                                          &c6_00,&c12_00);
 358
 359             /* REACTION-FIELD ELECTROSTATICS */
 360             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 361             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 362
 363             /* LENNARD-JONES DISPERSION/REPULSION */
 364
 365             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 366             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 367             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 368             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 369             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 370
 371             d                = _mm_sub_ps(r00,rswitch);
 372             d                = _mm_max_ps(d,_mm_setzero_ps());
 373             d2               = _mm_mul_ps(d,d);
 374             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)))))));
 375
 376             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 377
 378             /* Evaluate switch function */
 379             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 380             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 381             vvdw             = _mm_mul_ps(vvdw,sw);
 382             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm_and_ps(velec,cutoff_mask);
 386             velec            = _mm_andnot_ps(dummy_mask,velec);
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 389             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 390             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 391
 392             fscal            = _mm_add_ps(felec,fvdw);
 393
 394             fscal            = _mm_and_ps(fscal,cutoff_mask);
 395
 396             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx00);
 400             ty               = _mm_mul_ps(fscal,dy00);
 401             tz               = _mm_mul_ps(fscal,dz00);
 402
 403             /* Update vectorial force */
 404             fix0             = _mm_add_ps(fix0,tx);
 405             fiy0             = _mm_add_ps(fiy0,ty);
 406             fiz0             = _mm_add_ps(fiz0,tz);
 407
 408             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 409             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 410             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 411             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 413
 414             }
 415
 416             /* Inner loop uses 71 flops */
 417         }
 418
 419         /* End of innermost loop */
 420
 421         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 422                                               f+i_coord_offset,fshift+i_shift_offset);
 423
 424         ggid                        = gid[iidx];
 425         /* Update potential energies */
 426         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 427         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 428
 429         /* Increment number of inner iterations */
 430         inneriter                  += j_index_end - j_index_start;
 431
 432         /* Outer loop uses 9 flops */
 433     }
 434
 435     /* Increment number of outer iterations */
 436     outeriter        += nri;
 437
 438     /* Update outer/inner flops */
 439
 440     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*71);
 441 }
 442 /*
 443  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_single
 444  * Electrostatics interaction: ReactionField
 445  * VdW interaction:            LennardJones
 446  * Geometry:                   Particle-Particle
 447  * Calculate force/pot:        Force
 448  */
 449 void
 450 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_single
 451                     (t_nblist                    * gmx_restrict       nlist,
 452                      rvec                        * gmx_restrict          xx,
 453                      rvec                        * gmx_restrict          ff,
 454                      struct t_forcerec           * gmx_restrict          fr,
 455                      t_mdatoms                   * gmx_restrict     mdatoms,
 456                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 457                      t_nrnb                      * gmx_restrict        nrnb)
 458 {
 459     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 460      * just 0 for non-waters.
 461      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 462      * jnr indices corresponding to data put in the four positions in the SIMD register.
 463      */
 464     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 465     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 466     int              jnrA,jnrB,jnrC,jnrD;
 467     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 468     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 469     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 470     real             rcutoff_scalar;
 471     real             *shiftvec,*fshift,*x,*f;
 472     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 473     real             scratch[4*DIM];
 474     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 475     int              vdwioffset0;
 476     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 477     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 478     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 479     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 480     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 481     real             *charge;
 482     int              nvdwtype;
 483     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 484     int              *vdwtype;
 485     real             *vdwparam;
 486     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 487     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 488     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 489     real             rswitch_scalar,d_scalar;
 490     __m128           dummy_mask,cutoff_mask;
 491     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 492     __m128           one     = _mm_set1_ps(1.0);
 493     __m128           two     = _mm_set1_ps(2.0);
 494     x                = xx[0];
 495     f                = ff[0];
 496
 497     nri              = nlist->nri;
 498     iinr             = nlist->iinr;
 499     jindex           = nlist->jindex;
 500     jjnr             = nlist->jjnr;
 501     shiftidx         = nlist->shift;
 502     gid              = nlist->gid;
 503     shiftvec         = fr->shift_vec[0];
 504     fshift           = fr->fshift[0];
 505     facel            = _mm_set1_ps(fr->ic->epsfac);
 506     charge           = mdatoms->chargeA;
 507     krf              = _mm_set1_ps(fr->ic->k_rf);
 508     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 509     crf              = _mm_set1_ps(fr->ic->c_rf);
 510     nvdwtype         = fr->ntype;
 511     vdwparam         = fr->nbfp;
 512     vdwtype          = mdatoms->typeA;
 513
 514     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 515     rcutoff_scalar   = fr->ic->rcoulomb;
 516     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 517     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 518
 519     rswitch_scalar   = fr->ic->rvdw_switch;
 520     rswitch          = _mm_set1_ps(rswitch_scalar);
 521     /* Setup switch parameters */
 522     d_scalar         = rcutoff_scalar-rswitch_scalar;
 523     d                = _mm_set1_ps(d_scalar);
 524     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 525     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 526     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 527     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 528     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 529     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 530
 531     /* Avoid stupid compiler warnings */
 532     jnrA = jnrB = jnrC = jnrD = 0;
 533     j_coord_offsetA = 0;
 534     j_coord_offsetB = 0;
 535     j_coord_offsetC = 0;
 536     j_coord_offsetD = 0;
 537
 538     outeriter        = 0;
 539     inneriter        = 0;
 540
 541     for(iidx=0;iidx<4*DIM;iidx++)
 542     {
 543         scratch[iidx] = 0.0;
 544     }
 545
 546     /* Start outer loop over neighborlists */
 547     for(iidx=0; iidx<nri; iidx++)
 548     {
 549         /* Load shift vector for this list */
 550         i_shift_offset   = DIM*shiftidx[iidx];
 551
 552         /* Load limits for loop over neighbors */
 553         j_index_start    = jindex[iidx];
 554         j_index_end      = jindex[iidx+1];
 555
 556         /* Get outer coordinate index */
 557         inr              = iinr[iidx];
 558         i_coord_offset   = DIM*inr;
 559
 560         /* Load i particle coords and add shift vector */
 561         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 562
 563         fix0             = _mm_setzero_ps();
 564         fiy0             = _mm_setzero_ps();
 565         fiz0             = _mm_setzero_ps();
 566
 567         /* Load parameters for i particles */
 568         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 569         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 570
 571         /* Start inner kernel loop */
 572         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 573         {
 574
 575             /* Get j neighbor index, and coordinate index */
 576             jnrA             = jjnr[jidx];
 577             jnrB             = jjnr[jidx+1];
 578             jnrC             = jjnr[jidx+2];
 579             jnrD             = jjnr[jidx+3];
 580             j_coord_offsetA  = DIM*jnrA;
 581             j_coord_offsetB  = DIM*jnrB;
 582             j_coord_offsetC  = DIM*jnrC;
 583             j_coord_offsetD  = DIM*jnrD;
 584
 585             /* load j atom coordinates */
 586             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 587                                               x+j_coord_offsetC,x+j_coord_offsetD,
 588                                               &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _mm_sub_ps(ix0,jx0);
 592             dy00             = _mm_sub_ps(iy0,jy0);
 593             dz00             = _mm_sub_ps(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 597
 598             rinv00           = sse41_invsqrt_f(rsq00);
 599
 600             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 601
 602             /* Load parameters for j particles */
 603             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 604                                                               charge+jnrC+0,charge+jnrD+0);
 605             vdwjidx0A        = 2*vdwtype[jnrA+0];
 606             vdwjidx0B        = 2*vdwtype[jnrB+0];
 607             vdwjidx0C        = 2*vdwtype[jnrC+0];
 608             vdwjidx0D        = 2*vdwtype[jnrD+0];
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm_any_lt(rsq00,rcutoff2))
 615             {
 616
 617             r00              = _mm_mul_ps(rsq00,rinv00);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq00             = _mm_mul_ps(iq0,jq0);
 621             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 622                                          vdwparam+vdwioffset0+vdwjidx0B,
 623                                          vdwparam+vdwioffset0+vdwjidx0C,
 624                                          vdwparam+vdwioffset0+vdwjidx0D,
 625                                          &c6_00,&c12_00);
 626
 627             /* REACTION-FIELD ELECTROSTATICS */
 628             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 629
 630             /* LENNARD-JONES DISPERSION/REPULSION */
 631
 632             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 633             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 634             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 635             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 636             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 637
 638             d                = _mm_sub_ps(r00,rswitch);
 639             d                = _mm_max_ps(d,_mm_setzero_ps());
 640             d2               = _mm_mul_ps(d,d);
 641             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)))))));
 642
 643             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 644
 645             /* Evaluate switch function */
 646             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 647             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 648             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 649
 650             fscal            = _mm_add_ps(felec,fvdw);
 651
 652             fscal            = _mm_and_ps(fscal,cutoff_mask);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm_mul_ps(fscal,dx00);
 656             ty               = _mm_mul_ps(fscal,dy00);
 657             tz               = _mm_mul_ps(fscal,dz00);
 658
 659             /* Update vectorial force */
 660             fix0             = _mm_add_ps(fix0,tx);
 661             fiy0             = _mm_add_ps(fiy0,ty);
 662             fiz0             = _mm_add_ps(fiz0,tz);
 663
 664             fjptrA             = f+j_coord_offsetA;
 665             fjptrB             = f+j_coord_offsetB;
 666             fjptrC             = f+j_coord_offsetC;
 667             fjptrD             = f+j_coord_offsetD;
 668             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 669
 670             }
 671
 672             /* Inner loop uses 61 flops */
 673         }
 674
 675         if(jidx<j_index_end)
 676         {
 677
 678             /* Get j neighbor index, and coordinate index */
 679             jnrlistA         = jjnr[jidx];
 680             jnrlistB         = jjnr[jidx+1];
 681             jnrlistC         = jjnr[jidx+2];
 682             jnrlistD         = jjnr[jidx+3];
 683             /* Sign of each element will be negative for non-real atoms.
 684              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 685              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 686              */
 687             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 688             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 689             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 690             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 691             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 692             j_coord_offsetA  = DIM*jnrA;
 693             j_coord_offsetB  = DIM*jnrB;
 694             j_coord_offsetC  = DIM*jnrC;
 695             j_coord_offsetD  = DIM*jnrD;
 696
 697             /* load j atom coordinates */
 698             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 699                                               x+j_coord_offsetC,x+j_coord_offsetD,
 700                                               &jx0,&jy0,&jz0);
 701
 702             /* Calculate displacement vector */
 703             dx00             = _mm_sub_ps(ix0,jx0);
 704             dy00             = _mm_sub_ps(iy0,jy0);
 705             dz00             = _mm_sub_ps(iz0,jz0);
 706
 707             /* Calculate squared distance and things based on it */
 708             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 709
 710             rinv00           = sse41_invsqrt_f(rsq00);
 711
 712             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 713
 714             /* Load parameters for j particles */
 715             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 716                                                               charge+jnrC+0,charge+jnrD+0);
 717             vdwjidx0A        = 2*vdwtype[jnrA+0];
 718             vdwjidx0B        = 2*vdwtype[jnrB+0];
 719             vdwjidx0C        = 2*vdwtype[jnrC+0];
 720             vdwjidx0D        = 2*vdwtype[jnrD+0];
 721
 722             /**************************
 723              * CALCULATE INTERACTIONS *
 724              **************************/
 725
 726             if (gmx_mm_any_lt(rsq00,rcutoff2))
 727             {
 728
 729             r00              = _mm_mul_ps(rsq00,rinv00);
 730             r00              = _mm_andnot_ps(dummy_mask,r00);
 731
 732             /* Compute parameters for interactions between i and j atoms */
 733             qq00             = _mm_mul_ps(iq0,jq0);
 734             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 735                                          vdwparam+vdwioffset0+vdwjidx0B,
 736                                          vdwparam+vdwioffset0+vdwjidx0C,
 737                                          vdwparam+vdwioffset0+vdwjidx0D,
 738                                          &c6_00,&c12_00);
 739
 740             /* REACTION-FIELD ELECTROSTATICS */
 741             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 742
 743             /* LENNARD-JONES DISPERSION/REPULSION */
 744
 745             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 746             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 747             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 748             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 749             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 750
 751             d                = _mm_sub_ps(r00,rswitch);
 752             d                = _mm_max_ps(d,_mm_setzero_ps());
 753             d2               = _mm_mul_ps(d,d);
 754             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)))))));
 755
 756             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 757
 758             /* Evaluate switch function */
 759             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 760             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 761             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 762
 763             fscal            = _mm_add_ps(felec,fvdw);
 764
 765             fscal            = _mm_and_ps(fscal,cutoff_mask);
 766
 767             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 768
 769             /* Calculate temporary vectorial force */
 770             tx               = _mm_mul_ps(fscal,dx00);
 771             ty               = _mm_mul_ps(fscal,dy00);
 772             tz               = _mm_mul_ps(fscal,dz00);
 773
 774             /* Update vectorial force */
 775             fix0             = _mm_add_ps(fix0,tx);
 776             fiy0             = _mm_add_ps(fiy0,ty);
 777             fiz0             = _mm_add_ps(fiz0,tz);
 778
 779             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 780             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 781             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 782             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 783             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 784
 785             }
 786
 787             /* Inner loop uses 62 flops */
 788         }
 789
 790         /* End of innermost loop */
 791
 792         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 793                                               f+i_coord_offset,fshift+i_shift_offset);
 794
 795         /* Increment number of inner iterations */
 796         inneriter                  += j_index_end - j_index_start;
 797
 798         /* Outer loop uses 7 flops */
 799     }
 800
 801     /* Increment number of outer iterations */
 802     outeriter        += nri;
 803
 804     /* Update outer/inner flops */
 805
 806     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 807 }