src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 #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_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_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     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->ic->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_elec->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_ps();
 156         fiy0             = _mm_setzero_ps();
 157         fiz0             = _mm_setzero_ps();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_ps();
 165         vvdwsum          = _mm_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               x+j_coord_offsetC,x+j_coord_offsetD,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_ps(ix0,jx0);
 188             dy00             = _mm_sub_ps(iy0,jy0);
 189             dz00             = _mm_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = sse2_invsqrt_f(rsq00);
 195
 196             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203             vdwjidx0C        = 2*vdwtype[jnrC+0];
 204             vdwjidx0D        = 2*vdwtype[jnrD+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r00              = _mm_mul_ps(rsq00,rinv00);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq00             = _mm_mul_ps(iq0,jq0);
 214             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 215                                          vdwparam+vdwioffset0+vdwjidx0B,
 216                                          vdwparam+vdwioffset0+vdwjidx0C,
 217                                          vdwparam+vdwioffset0+vdwjidx0D,
 218                                          &c6_00,&c12_00);
 219
 220             /* Calculate table index by multiplying r with table scale and truncate to integer */
 221             rt               = _mm_mul_ps(r00,vftabscale);
 222             vfitab           = _mm_cvttps_epi32(rt);
 223             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 224             vfitab           = _mm_slli_epi32(vfitab,2);
 225
 226             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 227             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 228             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 229             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 230             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 231             _MM_TRANSPOSE4_PS(Y,F,G,H);
 232             Heps             = _mm_mul_ps(vfeps,H);
 233             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 234             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 235             velec            = _mm_mul_ps(qq00,VV);
 236             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 237             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 238
 239             /* LENNARD-JONES DISPERSION/REPULSION */
 240
 241             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 242             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 243             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 244             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 245             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_ps(felec,fvdw);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             fjptrA             = f+j_coord_offsetA;
 264             fjptrB             = f+j_coord_offsetB;
 265             fjptrC             = f+j_coord_offsetC;
 266             fjptrD             = f+j_coord_offsetD;
 267             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 268
 269             /* Inner loop uses 56 flops */
 270         }
 271
 272         if(jidx<j_index_end)
 273         {
 274
 275             /* Get j neighbor index, and coordinate index */
 276             jnrlistA         = jjnr[jidx];
 277             jnrlistB         = jjnr[jidx+1];
 278             jnrlistC         = jjnr[jidx+2];
 279             jnrlistD         = jjnr[jidx+3];
 280             /* Sign of each element will be negative for non-real atoms.
 281              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 282              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 283              */
 284             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 285             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 286             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 287             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 288             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 289             j_coord_offsetA  = DIM*jnrA;
 290             j_coord_offsetB  = DIM*jnrB;
 291             j_coord_offsetC  = DIM*jnrC;
 292             j_coord_offsetD  = DIM*jnrD;
 293
 294             /* load j atom coordinates */
 295             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 296                                               x+j_coord_offsetC,x+j_coord_offsetD,
 297                                               &jx0,&jy0,&jz0);
 298
 299             /* Calculate displacement vector */
 300             dx00             = _mm_sub_ps(ix0,jx0);
 301             dy00             = _mm_sub_ps(iy0,jy0);
 302             dz00             = _mm_sub_ps(iz0,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 306
 307             rinv00           = sse2_invsqrt_f(rsq00);
 308
 309             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 310
 311             /* Load parameters for j particles */
 312             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 313                                                               charge+jnrC+0,charge+jnrD+0);
 314             vdwjidx0A        = 2*vdwtype[jnrA+0];
 315             vdwjidx0B        = 2*vdwtype[jnrB+0];
 316             vdwjidx0C        = 2*vdwtype[jnrC+0];
 317             vdwjidx0D        = 2*vdwtype[jnrD+0];
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r00              = _mm_mul_ps(rsq00,rinv00);
 324             r00              = _mm_andnot_ps(dummy_mask,r00);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq00             = _mm_mul_ps(iq0,jq0);
 328             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 329                                          vdwparam+vdwioffset0+vdwjidx0B,
 330                                          vdwparam+vdwioffset0+vdwjidx0C,
 331                                          vdwparam+vdwioffset0+vdwjidx0D,
 332                                          &c6_00,&c12_00);
 333
 334             /* Calculate table index by multiplying r with table scale and truncate to integer */
 335             rt               = _mm_mul_ps(r00,vftabscale);
 336             vfitab           = _mm_cvttps_epi32(rt);
 337             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 338             vfitab           = _mm_slli_epi32(vfitab,2);
 339
 340             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 341             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 342             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 343             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 344             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 345             _MM_TRANSPOSE4_PS(Y,F,G,H);
 346             Heps             = _mm_mul_ps(vfeps,H);
 347             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 348             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 349             velec            = _mm_mul_ps(qq00,VV);
 350             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 351             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 352
 353             /* LENNARD-JONES DISPERSION/REPULSION */
 354
 355             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 356             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 357             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 358             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 359             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_andnot_ps(dummy_mask,velec);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 365             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 366
 367             fscal            = _mm_add_ps(felec,fvdw);
 368
 369             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx00);
 373             ty               = _mm_mul_ps(fscal,dy00);
 374             tz               = _mm_mul_ps(fscal,dz00);
 375
 376             /* Update vectorial force */
 377             fix0             = _mm_add_ps(fix0,tx);
 378             fiy0             = _mm_add_ps(fiy0,ty);
 379             fiz0             = _mm_add_ps(fiz0,tz);
 380
 381             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 382             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 383             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 384             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 385             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 386
 387             /* Inner loop uses 57 flops */
 388         }
 389
 390         /* End of innermost loop */
 391
 392         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 393                                               f+i_coord_offset,fshift+i_shift_offset);
 394
 395         ggid                        = gid[iidx];
 396         /* Update potential energies */
 397         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 398         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 399
 400         /* Increment number of inner iterations */
 401         inneriter                  += j_index_end - j_index_start;
 402
 403         /* Outer loop uses 9 flops */
 404     }
 405
 406     /* Increment number of outer iterations */
 407     outeriter        += nri;
 408
 409     /* Update outer/inner flops */
 410
 411     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 412 }
 413 /*
 414  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 415  * Electrostatics interaction: CubicSplineTable
 416  * VdW interaction:            LennardJones
 417  * Geometry:                   Particle-Particle
 418  * Calculate force/pot:        Force
 419  */
 420 void
 421 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 422                     (t_nblist                    * gmx_restrict       nlist,
 423                      rvec                        * gmx_restrict          xx,
 424                      rvec                        * gmx_restrict          ff,
 425                      struct t_forcerec           * gmx_restrict          fr,
 426                      t_mdatoms                   * gmx_restrict     mdatoms,
 427                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 428                      t_nrnb                      * gmx_restrict        nrnb)
 429 {
 430     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 431      * just 0 for non-waters.
 432      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 433      * jnr indices corresponding to data put in the four positions in the SIMD register.
 434      */
 435     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 436     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 437     int              jnrA,jnrB,jnrC,jnrD;
 438     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 439     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 440     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 441     real             rcutoff_scalar;
 442     real             *shiftvec,*fshift,*x,*f;
 443     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 444     real             scratch[4*DIM];
 445     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 446     int              vdwioffset0;
 447     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 448     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 449     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 450     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 451     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 452     real             *charge;
 453     int              nvdwtype;
 454     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 455     int              *vdwtype;
 456     real             *vdwparam;
 457     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 458     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 459     __m128i          vfitab;
 460     __m128i          ifour       = _mm_set1_epi32(4);
 461     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 462     real             *vftab;
 463     __m128           dummy_mask,cutoff_mask;
 464     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 465     __m128           one     = _mm_set1_ps(1.0);
 466     __m128           two     = _mm_set1_ps(2.0);
 467     x                = xx[0];
 468     f                = ff[0];
 469
 470     nri              = nlist->nri;
 471     iinr             = nlist->iinr;
 472     jindex           = nlist->jindex;
 473     jjnr             = nlist->jjnr;
 474     shiftidx         = nlist->shift;
 475     gid              = nlist->gid;
 476     shiftvec         = fr->shift_vec[0];
 477     fshift           = fr->fshift[0];
 478     facel            = _mm_set1_ps(fr->ic->epsfac);
 479     charge           = mdatoms->chargeA;
 480     nvdwtype         = fr->ntype;
 481     vdwparam         = fr->nbfp;
 482     vdwtype          = mdatoms->typeA;
 483
 484     vftab            = kernel_data->table_elec->data;
 485     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 486
 487     /* Avoid stupid compiler warnings */
 488     jnrA = jnrB = jnrC = jnrD = 0;
 489     j_coord_offsetA = 0;
 490     j_coord_offsetB = 0;
 491     j_coord_offsetC = 0;
 492     j_coord_offsetD = 0;
 493
 494     outeriter        = 0;
 495     inneriter        = 0;
 496
 497     for(iidx=0;iidx<4*DIM;iidx++)
 498     {
 499         scratch[iidx] = 0.0;
 500     }
 501
 502     /* Start outer loop over neighborlists */
 503     for(iidx=0; iidx<nri; iidx++)
 504     {
 505         /* Load shift vector for this list */
 506         i_shift_offset   = DIM*shiftidx[iidx];
 507
 508         /* Load limits for loop over neighbors */
 509         j_index_start    = jindex[iidx];
 510         j_index_end      = jindex[iidx+1];
 511
 512         /* Get outer coordinate index */
 513         inr              = iinr[iidx];
 514         i_coord_offset   = DIM*inr;
 515
 516         /* Load i particle coords and add shift vector */
 517         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 518
 519         fix0             = _mm_setzero_ps();
 520         fiy0             = _mm_setzero_ps();
 521         fiz0             = _mm_setzero_ps();
 522
 523         /* Load parameters for i particles */
 524         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 525         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 526
 527         /* Start inner kernel loop */
 528         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 529         {
 530
 531             /* Get j neighbor index, and coordinate index */
 532             jnrA             = jjnr[jidx];
 533             jnrB             = jjnr[jidx+1];
 534             jnrC             = jjnr[jidx+2];
 535             jnrD             = jjnr[jidx+3];
 536             j_coord_offsetA  = DIM*jnrA;
 537             j_coord_offsetB  = DIM*jnrB;
 538             j_coord_offsetC  = DIM*jnrC;
 539             j_coord_offsetD  = DIM*jnrD;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 543                                               x+j_coord_offsetC,x+j_coord_offsetD,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _mm_sub_ps(ix0,jx0);
 548             dy00             = _mm_sub_ps(iy0,jy0);
 549             dz00             = _mm_sub_ps(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 553
 554             rinv00           = sse2_invsqrt_f(rsq00);
 555
 556             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 557
 558             /* Load parameters for j particles */
 559             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 560                                                               charge+jnrC+0,charge+jnrD+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562             vdwjidx0B        = 2*vdwtype[jnrB+0];
 563             vdwjidx0C        = 2*vdwtype[jnrC+0];
 564             vdwjidx0D        = 2*vdwtype[jnrD+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r00              = _mm_mul_ps(rsq00,rinv00);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq00             = _mm_mul_ps(iq0,jq0);
 574             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 575                                          vdwparam+vdwioffset0+vdwjidx0B,
 576                                          vdwparam+vdwioffset0+vdwjidx0C,
 577                                          vdwparam+vdwioffset0+vdwjidx0D,
 578                                          &c6_00,&c12_00);
 579
 580             /* Calculate table index by multiplying r with table scale and truncate to integer */
 581             rt               = _mm_mul_ps(r00,vftabscale);
 582             vfitab           = _mm_cvttps_epi32(rt);
 583             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 584             vfitab           = _mm_slli_epi32(vfitab,2);
 585
 586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 587             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 588             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 589             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 590             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 591             _MM_TRANSPOSE4_PS(Y,F,G,H);
 592             Heps             = _mm_mul_ps(vfeps,H);
 593             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 594             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 595             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 596
 597             /* LENNARD-JONES DISPERSION/REPULSION */
 598
 599             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 600             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 601
 602             fscal            = _mm_add_ps(felec,fvdw);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm_mul_ps(fscal,dx00);
 606             ty               = _mm_mul_ps(fscal,dy00);
 607             tz               = _mm_mul_ps(fscal,dz00);
 608
 609             /* Update vectorial force */
 610             fix0             = _mm_add_ps(fix0,tx);
 611             fiy0             = _mm_add_ps(fiy0,ty);
 612             fiz0             = _mm_add_ps(fiz0,tz);
 613
 614             fjptrA             = f+j_coord_offsetA;
 615             fjptrB             = f+j_coord_offsetB;
 616             fjptrC             = f+j_coord_offsetC;
 617             fjptrD             = f+j_coord_offsetD;
 618             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 619
 620             /* Inner loop uses 47 flops */
 621         }
 622
 623         if(jidx<j_index_end)
 624         {
 625
 626             /* Get j neighbor index, and coordinate index */
 627             jnrlistA         = jjnr[jidx];
 628             jnrlistB         = jjnr[jidx+1];
 629             jnrlistC         = jjnr[jidx+2];
 630             jnrlistD         = jjnr[jidx+3];
 631             /* Sign of each element will be negative for non-real atoms.
 632              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 633              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 634              */
 635             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 636             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 637             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 638             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 639             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 640             j_coord_offsetA  = DIM*jnrA;
 641             j_coord_offsetB  = DIM*jnrB;
 642             j_coord_offsetC  = DIM*jnrC;
 643             j_coord_offsetD  = DIM*jnrD;
 644
 645             /* load j atom coordinates */
 646             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 647                                               x+j_coord_offsetC,x+j_coord_offsetD,
 648                                               &jx0,&jy0,&jz0);
 649
 650             /* Calculate displacement vector */
 651             dx00             = _mm_sub_ps(ix0,jx0);
 652             dy00             = _mm_sub_ps(iy0,jy0);
 653             dz00             = _mm_sub_ps(iz0,jz0);
 654
 655             /* Calculate squared distance and things based on it */
 656             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 657
 658             rinv00           = sse2_invsqrt_f(rsq00);
 659
 660             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 661
 662             /* Load parameters for j particles */
 663             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 664                                                               charge+jnrC+0,charge+jnrD+0);
 665             vdwjidx0A        = 2*vdwtype[jnrA+0];
 666             vdwjidx0B        = 2*vdwtype[jnrB+0];
 667             vdwjidx0C        = 2*vdwtype[jnrC+0];
 668             vdwjidx0D        = 2*vdwtype[jnrD+0];
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             r00              = _mm_mul_ps(rsq00,rinv00);
 675             r00              = _mm_andnot_ps(dummy_mask,r00);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq00             = _mm_mul_ps(iq0,jq0);
 679             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 680                                          vdwparam+vdwioffset0+vdwjidx0B,
 681                                          vdwparam+vdwioffset0+vdwjidx0C,
 682                                          vdwparam+vdwioffset0+vdwjidx0D,
 683                                          &c6_00,&c12_00);
 684
 685             /* Calculate table index by multiplying r with table scale and truncate to integer */
 686             rt               = _mm_mul_ps(r00,vftabscale);
 687             vfitab           = _mm_cvttps_epi32(rt);
 688             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 689             vfitab           = _mm_slli_epi32(vfitab,2);
 690
 691             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 692             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 693             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 694             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 695             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 696             _MM_TRANSPOSE4_PS(Y,F,G,H);
 697             Heps             = _mm_mul_ps(vfeps,H);
 698             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 699             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 700             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 701
 702             /* LENNARD-JONES DISPERSION/REPULSION */
 703
 704             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 705             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 706
 707             fscal            = _mm_add_ps(felec,fvdw);
 708
 709             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm_mul_ps(fscal,dx00);
 713             ty               = _mm_mul_ps(fscal,dy00);
 714             tz               = _mm_mul_ps(fscal,dz00);
 715
 716             /* Update vectorial force */
 717             fix0             = _mm_add_ps(fix0,tx);
 718             fiy0             = _mm_add_ps(fiy0,ty);
 719             fiz0             = _mm_add_ps(fiz0,tz);
 720
 721             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 722             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 723             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 724             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 725             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 726
 727             /* Inner loop uses 48 flops */
 728         }
 729
 730         /* End of innermost loop */
 731
 732         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 733                                               f+i_coord_offset,fshift+i_shift_offset);
 734
 735         /* Increment number of inner iterations */
 736         inneriter                  += j_index_end - j_index_start;
 737
 738         /* Outer loop uses 7 flops */
 739     }
 740
 741     /* Increment number of outer iterations */
 742     outeriter        += nri;
 743
 744     /* Update outer/inner flops */
 745
 746     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 747 }