src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwLJ_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_elec->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm_setzero_ps();
 159         fiy0             = _mm_setzero_ps();
 160         fiz0             = _mm_setzero_ps();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 164         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 196
 197             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 198
 199             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                               charge+jnrC+0,charge+jnrD+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206             vdwjidx0C        = 2*vdwtype[jnrC+0];
 207             vdwjidx0D        = 2*vdwtype[jnrD+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             r00              = _mm_mul_ps(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_ps(iq0,jq0);
 217             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,
 219                                          vdwparam+vdwioffset0+vdwjidx0C,
 220                                          vdwparam+vdwioffset0+vdwjidx0D,
 221                                          &c6_00,&c12_00);
 222
 223             /* Calculate table index by multiplying r with table scale and truncate to integer */
 224             rt               = _mm_mul_ps(r00,vftabscale);
 225             vfitab           = _mm_cvttps_epi32(rt);
 226             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 227             vfitab           = _mm_slli_epi32(vfitab,2);
 228
 229             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 230             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 231             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 232             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 233             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 234             _MM_TRANSPOSE4_PS(Y,F,G,H);
 235             Heps             = _mm_mul_ps(vfeps,H);
 236             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 237             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 238             velec            = _mm_mul_ps(qq00,VV);
 239             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 240             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 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             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm_add_ps(velecsum,velec);
 252             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 253
 254             fscal            = _mm_add_ps(felec,fvdw);
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_ps(fscal,dx00);
 258             ty               = _mm_mul_ps(fscal,dy00);
 259             tz               = _mm_mul_ps(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_ps(fix0,tx);
 263             fiy0             = _mm_add_ps(fiy0,ty);
 264             fiz0             = _mm_add_ps(fiz0,tz);
 265
 266             fjptrA             = f+j_coord_offsetA;
 267             fjptrB             = f+j_coord_offsetB;
 268             fjptrC             = f+j_coord_offsetC;
 269             fjptrD             = f+j_coord_offsetD;
 270             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 271
 272             /* Inner loop uses 56 flops */
 273         }
 274
 275         if(jidx<j_index_end)
 276         {
 277
 278             /* Get j neighbor index, and coordinate index */
 279             jnrlistA         = jjnr[jidx];
 280             jnrlistB         = jjnr[jidx+1];
 281             jnrlistC         = jjnr[jidx+2];
 282             jnrlistD         = jjnr[jidx+3];
 283             /* Sign of each element will be negative for non-real atoms.
 284              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 285              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 286              */
 287             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 288             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 289             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 290             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 291             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 292             j_coord_offsetA  = DIM*jnrA;
 293             j_coord_offsetB  = DIM*jnrB;
 294             j_coord_offsetC  = DIM*jnrC;
 295             j_coord_offsetD  = DIM*jnrD;
 296
 297             /* load j atom coordinates */
 298             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 299                                               x+j_coord_offsetC,x+j_coord_offsetD,
 300                                               &jx0,&jy0,&jz0);
 301
 302             /* Calculate displacement vector */
 303             dx00             = _mm_sub_ps(ix0,jx0);
 304             dy00             = _mm_sub_ps(iy0,jy0);
 305             dz00             = _mm_sub_ps(iz0,jz0);
 306
 307             /* Calculate squared distance and things based on it */
 308             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 309
 310             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 311
 312             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 313
 314             /* Load parameters for j particles */
 315             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 316                                                               charge+jnrC+0,charge+jnrD+0);
 317             vdwjidx0A        = 2*vdwtype[jnrA+0];
 318             vdwjidx0B        = 2*vdwtype[jnrB+0];
 319             vdwjidx0C        = 2*vdwtype[jnrC+0];
 320             vdwjidx0D        = 2*vdwtype[jnrD+0];
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r00              = _mm_mul_ps(rsq00,rinv00);
 327             r00              = _mm_andnot_ps(dummy_mask,r00);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq00             = _mm_mul_ps(iq0,jq0);
 331             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 332                                          vdwparam+vdwioffset0+vdwjidx0B,
 333                                          vdwparam+vdwioffset0+vdwjidx0C,
 334                                          vdwparam+vdwioffset0+vdwjidx0D,
 335                                          &c6_00,&c12_00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm_mul_ps(r00,vftabscale);
 339             vfitab           = _mm_cvttps_epi32(rt);
 340             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 341             vfitab           = _mm_slli_epi32(vfitab,2);
 342
 343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 344             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 345             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 346             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 347             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 348             _MM_TRANSPOSE4_PS(Y,F,G,H);
 349             Heps             = _mm_mul_ps(vfeps,H);
 350             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 351             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 352             velec            = _mm_mul_ps(qq00,VV);
 353             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 354             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 355
 356             /* LENNARD-JONES DISPERSION/REPULSION */
 357
 358             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 359             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 360             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 361             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 362             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_andnot_ps(dummy_mask,velec);
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 368             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 369
 370             fscal            = _mm_add_ps(felec,fvdw);
 371
 372             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_ps(fscal,dx00);
 376             ty               = _mm_mul_ps(fscal,dy00);
 377             tz               = _mm_mul_ps(fscal,dz00);
 378
 379             /* Update vectorial force */
 380             fix0             = _mm_add_ps(fix0,tx);
 381             fiy0             = _mm_add_ps(fiy0,ty);
 382             fiz0             = _mm_add_ps(fiz0,tz);
 383
 384             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 385             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 386             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 387             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 388             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 389
 390             /* Inner loop uses 57 flops */
 391         }
 392
 393         /* End of innermost loop */
 394
 395         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 396                                               f+i_coord_offset,fshift+i_shift_offset);
 397
 398         ggid                        = gid[iidx];
 399         /* Update potential energies */
 400         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 401         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 402
 403         /* Increment number of inner iterations */
 404         inneriter                  += j_index_end - j_index_start;
 405
 406         /* Outer loop uses 9 flops */
 407     }
 408
 409     /* Increment number of outer iterations */
 410     outeriter        += nri;
 411
 412     /* Update outer/inner flops */
 413
 414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 415 }
 416 /*
 417  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse4_1_single
 418  * Electrostatics interaction: CubicSplineTable
 419  * VdW interaction:            LennardJones
 420  * Geometry:                   Particle-Particle
 421  * Calculate force/pot:        Force
 422  */
 423 void
 424 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse4_1_single
 425                     (t_nblist                    * gmx_restrict       nlist,
 426                      rvec                        * gmx_restrict          xx,
 427                      rvec                        * gmx_restrict          ff,
 428                      t_forcerec                  * gmx_restrict          fr,
 429                      t_mdatoms                   * gmx_restrict     mdatoms,
 430                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 431                      t_nrnb                      * gmx_restrict        nrnb)
 432 {
 433     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 434      * just 0 for non-waters.
 435      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 436      * jnr indices corresponding to data put in the four positions in the SIMD register.
 437      */
 438     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 439     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 440     int              jnrA,jnrB,jnrC,jnrD;
 441     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 442     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 443     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 444     real             rcutoff_scalar;
 445     real             *shiftvec,*fshift,*x,*f;
 446     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 447     real             scratch[4*DIM];
 448     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 449     int              vdwioffset0;
 450     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 451     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 452     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 453     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 454     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 455     real             *charge;
 456     int              nvdwtype;
 457     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 458     int              *vdwtype;
 459     real             *vdwparam;
 460     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 461     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 462     __m128i          vfitab;
 463     __m128i          ifour       = _mm_set1_epi32(4);
 464     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 465     real             *vftab;
 466     __m128           dummy_mask,cutoff_mask;
 467     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 468     __m128           one     = _mm_set1_ps(1.0);
 469     __m128           two     = _mm_set1_ps(2.0);
 470     x                = xx[0];
 471     f                = ff[0];
 472
 473     nri              = nlist->nri;
 474     iinr             = nlist->iinr;
 475     jindex           = nlist->jindex;
 476     jjnr             = nlist->jjnr;
 477     shiftidx         = nlist->shift;
 478     gid              = nlist->gid;
 479     shiftvec         = fr->shift_vec[0];
 480     fshift           = fr->fshift[0];
 481     facel            = _mm_set1_ps(fr->epsfac);
 482     charge           = mdatoms->chargeA;
 483     nvdwtype         = fr->ntype;
 484     vdwparam         = fr->nbfp;
 485     vdwtype          = mdatoms->typeA;
 486
 487     vftab            = kernel_data->table_elec->data;
 488     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 489
 490     /* Avoid stupid compiler warnings */
 491     jnrA = jnrB = jnrC = jnrD = 0;
 492     j_coord_offsetA = 0;
 493     j_coord_offsetB = 0;
 494     j_coord_offsetC = 0;
 495     j_coord_offsetD = 0;
 496
 497     outeriter        = 0;
 498     inneriter        = 0;
 499
 500     for(iidx=0;iidx<4*DIM;iidx++)
 501     {
 502         scratch[iidx] = 0.0;
 503     }
 504
 505     /* Start outer loop over neighborlists */
 506     for(iidx=0; iidx<nri; iidx++)
 507     {
 508         /* Load shift vector for this list */
 509         i_shift_offset   = DIM*shiftidx[iidx];
 510
 511         /* Load limits for loop over neighbors */
 512         j_index_start    = jindex[iidx];
 513         j_index_end      = jindex[iidx+1];
 514
 515         /* Get outer coordinate index */
 516         inr              = iinr[iidx];
 517         i_coord_offset   = DIM*inr;
 518
 519         /* Load i particle coords and add shift vector */
 520         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 521
 522         fix0             = _mm_setzero_ps();
 523         fiy0             = _mm_setzero_ps();
 524         fiz0             = _mm_setzero_ps();
 525
 526         /* Load parameters for i particles */
 527         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 528         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 529
 530         /* Start inner kernel loop */
 531         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 532         {
 533
 534             /* Get j neighbor index, and coordinate index */
 535             jnrA             = jjnr[jidx];
 536             jnrB             = jjnr[jidx+1];
 537             jnrC             = jjnr[jidx+2];
 538             jnrD             = jjnr[jidx+3];
 539             j_coord_offsetA  = DIM*jnrA;
 540             j_coord_offsetB  = DIM*jnrB;
 541             j_coord_offsetC  = DIM*jnrC;
 542             j_coord_offsetD  = DIM*jnrD;
 543
 544             /* load j atom coordinates */
 545             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 546                                               x+j_coord_offsetC,x+j_coord_offsetD,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_ps(ix0,jx0);
 551             dy00             = _mm_sub_ps(iy0,jy0);
 552             dz00             = _mm_sub_ps(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 558
 559             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 563                                                               charge+jnrC+0,charge+jnrD+0);
 564             vdwjidx0A        = 2*vdwtype[jnrA+0];
 565             vdwjidx0B        = 2*vdwtype[jnrB+0];
 566             vdwjidx0C        = 2*vdwtype[jnrC+0];
 567             vdwjidx0D        = 2*vdwtype[jnrD+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r00              = _mm_mul_ps(rsq00,rinv00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq00             = _mm_mul_ps(iq0,jq0);
 577             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 578                                          vdwparam+vdwioffset0+vdwjidx0B,
 579                                          vdwparam+vdwioffset0+vdwjidx0C,
 580                                          vdwparam+vdwioffset0+vdwjidx0D,
 581                                          &c6_00,&c12_00);
 582
 583             /* Calculate table index by multiplying r with table scale and truncate to integer */
 584             rt               = _mm_mul_ps(r00,vftabscale);
 585             vfitab           = _mm_cvttps_epi32(rt);
 586             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 587             vfitab           = _mm_slli_epi32(vfitab,2);
 588
 589             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 590             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 591             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 592             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 593             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 594             _MM_TRANSPOSE4_PS(Y,F,G,H);
 595             Heps             = _mm_mul_ps(vfeps,H);
 596             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 597             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 598             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 599
 600             /* LENNARD-JONES DISPERSION/REPULSION */
 601
 602             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 603             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 604
 605             fscal            = _mm_add_ps(felec,fvdw);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm_mul_ps(fscal,dx00);
 609             ty               = _mm_mul_ps(fscal,dy00);
 610             tz               = _mm_mul_ps(fscal,dz00);
 611
 612             /* Update vectorial force */
 613             fix0             = _mm_add_ps(fix0,tx);
 614             fiy0             = _mm_add_ps(fiy0,ty);
 615             fiz0             = _mm_add_ps(fiz0,tz);
 616
 617             fjptrA             = f+j_coord_offsetA;
 618             fjptrB             = f+j_coord_offsetB;
 619             fjptrC             = f+j_coord_offsetC;
 620             fjptrD             = f+j_coord_offsetD;
 621             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 622
 623             /* Inner loop uses 47 flops */
 624         }
 625
 626         if(jidx<j_index_end)
 627         {
 628
 629             /* Get j neighbor index, and coordinate index */
 630             jnrlistA         = jjnr[jidx];
 631             jnrlistB         = jjnr[jidx+1];
 632             jnrlistC         = jjnr[jidx+2];
 633             jnrlistD         = jjnr[jidx+3];
 634             /* Sign of each element will be negative for non-real atoms.
 635              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 636              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 637              */
 638             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 639             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 640             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 641             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 642             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 643             j_coord_offsetA  = DIM*jnrA;
 644             j_coord_offsetB  = DIM*jnrB;
 645             j_coord_offsetC  = DIM*jnrC;
 646             j_coord_offsetD  = DIM*jnrD;
 647
 648             /* load j atom coordinates */
 649             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 650                                               x+j_coord_offsetC,x+j_coord_offsetD,
 651                                               &jx0,&jy0,&jz0);
 652
 653             /* Calculate displacement vector */
 654             dx00             = _mm_sub_ps(ix0,jx0);
 655             dy00             = _mm_sub_ps(iy0,jy0);
 656             dz00             = _mm_sub_ps(iz0,jz0);
 657
 658             /* Calculate squared distance and things based on it */
 659             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 660
 661             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 662
 663             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 664
 665             /* Load parameters for j particles */
 666             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 667                                                               charge+jnrC+0,charge+jnrD+0);
 668             vdwjidx0A        = 2*vdwtype[jnrA+0];
 669             vdwjidx0B        = 2*vdwtype[jnrB+0];
 670             vdwjidx0C        = 2*vdwtype[jnrC+0];
 671             vdwjidx0D        = 2*vdwtype[jnrD+0];
 672
 673             /**************************
 674              * CALCULATE INTERACTIONS *
 675              **************************/
 676
 677             r00              = _mm_mul_ps(rsq00,rinv00);
 678             r00              = _mm_andnot_ps(dummy_mask,r00);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq00             = _mm_mul_ps(iq0,jq0);
 682             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 683                                          vdwparam+vdwioffset0+vdwjidx0B,
 684                                          vdwparam+vdwioffset0+vdwjidx0C,
 685                                          vdwparam+vdwioffset0+vdwjidx0D,
 686                                          &c6_00,&c12_00);
 687
 688             /* Calculate table index by multiplying r with table scale and truncate to integer */
 689             rt               = _mm_mul_ps(r00,vftabscale);
 690             vfitab           = _mm_cvttps_epi32(rt);
 691             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 692             vfitab           = _mm_slli_epi32(vfitab,2);
 693
 694             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 695             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 696             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 697             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 698             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 699             _MM_TRANSPOSE4_PS(Y,F,G,H);
 700             Heps             = _mm_mul_ps(vfeps,H);
 701             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 702             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 703             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 704
 705             /* LENNARD-JONES DISPERSION/REPULSION */
 706
 707             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 708             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 709
 710             fscal            = _mm_add_ps(felec,fvdw);
 711
 712             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm_mul_ps(fscal,dx00);
 716             ty               = _mm_mul_ps(fscal,dy00);
 717             tz               = _mm_mul_ps(fscal,dz00);
 718
 719             /* Update vectorial force */
 720             fix0             = _mm_add_ps(fix0,tx);
 721             fiy0             = _mm_add_ps(fiy0,ty);
 722             fiz0             = _mm_add_ps(fiz0,tz);
 723
 724             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 725             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 726             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 727             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 728             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 729
 730             /* Inner loop uses 48 flops */
 731         }
 732
 733         /* End of innermost loop */
 734
 735         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 736                                               f+i_coord_offset,fshift+i_shift_offset);
 737
 738         /* Increment number of inner iterations */
 739         inneriter                  += j_index_end - j_index_start;
 740
 741         /* Outer loop uses 7 flops */
 742     }
 743
 744     /* Increment number of outer iterations */
 745     outeriter        += nri;
 746
 747     /* Update outer/inner flops */
 748
 749     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 750 }