src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwCSTab_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_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwCSTab_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     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_ps();
 155         fiy0             = _mm_setzero_ps();
 156         fiz0             = _mm_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         vvdwsum          = _mm_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               x+j_coord_offsetC,x+j_coord_offsetD,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_ps(ix0,jx0);
 185             dy00             = _mm_sub_ps(iy0,jy0);
 186             dz00             = _mm_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 192
 193             /* Load parameters for j particles */
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196             vdwjidx0C        = 2*vdwtype[jnrC+0];
 197             vdwjidx0D        = 2*vdwtype[jnrD+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _mm_mul_ps(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* Calculate table index by multiplying r with table scale and truncate to integer */
 213             rt               = _mm_mul_ps(r00,vftabscale);
 214             vfitab           = _mm_cvttps_epi32(rt);
 215             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 216             vfitab           = _mm_slli_epi32(vfitab,3);
 217
 218             /* CUBIC SPLINE TABLE DISPERSION */
 219             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 221             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 222             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 223             _MM_TRANSPOSE4_PS(Y,F,G,H);
 224             Heps             = _mm_mul_ps(vfeps,H);
 225             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 226             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 227             vvdw6            = _mm_mul_ps(c6_00,VV);
 228             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 229             fvdw6            = _mm_mul_ps(c6_00,FF);
 230
 231             /* CUBIC SPLINE TABLE REPULSION */
 232             vfitab           = _mm_add_epi32(vfitab,ifour);
 233             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 234             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 235             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 236             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 237             _MM_TRANSPOSE4_PS(Y,F,G,H);
 238             Heps             = _mm_mul_ps(vfeps,H);
 239             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 240             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 241             vvdw12           = _mm_mul_ps(c12_00,VV);
 242             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 243             fvdw12           = _mm_mul_ps(c12_00,FF);
 244             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 245             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 249
 250             fscal            = fvdw;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_ps(fscal,dx00);
 254             ty               = _mm_mul_ps(fscal,dy00);
 255             tz               = _mm_mul_ps(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_ps(fix0,tx);
 259             fiy0             = _mm_add_ps(fiy0,ty);
 260             fiz0             = _mm_add_ps(fiz0,tz);
 261
 262             fjptrA             = f+j_coord_offsetA;
 263             fjptrB             = f+j_coord_offsetB;
 264             fjptrC             = f+j_coord_offsetC;
 265             fjptrD             = f+j_coord_offsetD;
 266             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 267
 268             /* Inner loop uses 56 flops */
 269         }
 270
 271         if(jidx<j_index_end)
 272         {
 273
 274             /* Get j neighbor index, and coordinate index */
 275             jnrlistA         = jjnr[jidx];
 276             jnrlistB         = jjnr[jidx+1];
 277             jnrlistC         = jjnr[jidx+2];
 278             jnrlistD         = jjnr[jidx+3];
 279             /* Sign of each element will be negative for non-real atoms.
 280              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 281              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 282              */
 283             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 284             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 285             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 286             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 287             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 288             j_coord_offsetA  = DIM*jnrA;
 289             j_coord_offsetB  = DIM*jnrB;
 290             j_coord_offsetC  = DIM*jnrC;
 291             j_coord_offsetD  = DIM*jnrD;
 292
 293             /* load j atom coordinates */
 294             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 295                                               x+j_coord_offsetC,x+j_coord_offsetD,
 296                                               &jx0,&jy0,&jz0);
 297
 298             /* Calculate displacement vector */
 299             dx00             = _mm_sub_ps(ix0,jx0);
 300             dy00             = _mm_sub_ps(iy0,jy0);
 301             dz00             = _mm_sub_ps(iz0,jz0);
 302
 303             /* Calculate squared distance and things based on it */
 304             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 305
 306             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 307
 308             /* Load parameters for j particles */
 309             vdwjidx0A        = 2*vdwtype[jnrA+0];
 310             vdwjidx0B        = 2*vdwtype[jnrB+0];
 311             vdwjidx0C        = 2*vdwtype[jnrC+0];
 312             vdwjidx0D        = 2*vdwtype[jnrD+0];
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r00              = _mm_mul_ps(rsq00,rinv00);
 319             r00              = _mm_andnot_ps(dummy_mask,r00);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 323                                          vdwparam+vdwioffset0+vdwjidx0B,
 324                                          vdwparam+vdwioffset0+vdwjidx0C,
 325                                          vdwparam+vdwioffset0+vdwjidx0D,
 326                                          &c6_00,&c12_00);
 327
 328             /* Calculate table index by multiplying r with table scale and truncate to integer */
 329             rt               = _mm_mul_ps(r00,vftabscale);
 330             vfitab           = _mm_cvttps_epi32(rt);
 331             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 332             vfitab           = _mm_slli_epi32(vfitab,3);
 333
 334             /* CUBIC SPLINE TABLE DISPERSION */
 335             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 336             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 337             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 338             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 339             _MM_TRANSPOSE4_PS(Y,F,G,H);
 340             Heps             = _mm_mul_ps(vfeps,H);
 341             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 342             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 343             vvdw6            = _mm_mul_ps(c6_00,VV);
 344             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 345             fvdw6            = _mm_mul_ps(c6_00,FF);
 346
 347             /* CUBIC SPLINE TABLE REPULSION */
 348             vfitab           = _mm_add_epi32(vfitab,ifour);
 349             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 350             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 351             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 352             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 353             _MM_TRANSPOSE4_PS(Y,F,G,H);
 354             Heps             = _mm_mul_ps(vfeps,H);
 355             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 356             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 357             vvdw12           = _mm_mul_ps(c12_00,VV);
 358             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 359             fvdw12           = _mm_mul_ps(c12_00,FF);
 360             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 361             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 365             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 366
 367             fscal            = 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(vvdwsum,kernel_data->energygrp_vdw+ggid);
 398
 399         /* Increment number of inner iterations */
 400         inneriter                  += j_index_end - j_index_start;
 401
 402         /* Outer loop uses 7 flops */
 403     }
 404
 405     /* Increment number of outer iterations */
 406     outeriter        += nri;
 407
 408     /* Update outer/inner flops */
 409
 410     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*57);
 411 }
 412 /*
 413  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_single
 414  * Electrostatics interaction: None
 415  * VdW interaction:            CubicSplineTable
 416  * Geometry:                   Particle-Particle
 417  * Calculate force/pot:        Force
 418  */
 419 void
 420 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_single
 421                     (t_nblist                    * gmx_restrict       nlist,
 422                      rvec                        * gmx_restrict          xx,
 423                      rvec                        * gmx_restrict          ff,
 424                      t_forcerec                  * gmx_restrict          fr,
 425                      t_mdatoms                   * gmx_restrict     mdatoms,
 426                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 427                      t_nrnb                      * gmx_restrict        nrnb)
 428 {
 429     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 430      * just 0 for non-waters.
 431      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 432      * jnr indices corresponding to data put in the four positions in the SIMD register.
 433      */
 434     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 435     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 436     int              jnrA,jnrB,jnrC,jnrD;
 437     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 438     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 439     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 440     real             rcutoff_scalar;
 441     real             *shiftvec,*fshift,*x,*f;
 442     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 443     real             scratch[4*DIM];
 444     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 445     int              vdwioffset0;
 446     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 447     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 448     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 449     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 450     int              nvdwtype;
 451     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 452     int              *vdwtype;
 453     real             *vdwparam;
 454     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 455     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 456     __m128i          vfitab;
 457     __m128i          ifour       = _mm_set1_epi32(4);
 458     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 459     real             *vftab;
 460     __m128           dummy_mask,cutoff_mask;
 461     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 462     __m128           one     = _mm_set1_ps(1.0);
 463     __m128           two     = _mm_set1_ps(2.0);
 464     x                = xx[0];
 465     f                = ff[0];
 466
 467     nri              = nlist->nri;
 468     iinr             = nlist->iinr;
 469     jindex           = nlist->jindex;
 470     jjnr             = nlist->jjnr;
 471     shiftidx         = nlist->shift;
 472     gid              = nlist->gid;
 473     shiftvec         = fr->shift_vec[0];
 474     fshift           = fr->fshift[0];
 475     nvdwtype         = fr->ntype;
 476     vdwparam         = fr->nbfp;
 477     vdwtype          = mdatoms->typeA;
 478
 479     vftab            = kernel_data->table_vdw->data;
 480     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 481
 482     /* Avoid stupid compiler warnings */
 483     jnrA = jnrB = jnrC = jnrD = 0;
 484     j_coord_offsetA = 0;
 485     j_coord_offsetB = 0;
 486     j_coord_offsetC = 0;
 487     j_coord_offsetD = 0;
 488
 489     outeriter        = 0;
 490     inneriter        = 0;
 491
 492     for(iidx=0;iidx<4*DIM;iidx++)
 493     {
 494         scratch[iidx] = 0.0;
 495     }
 496
 497     /* Start outer loop over neighborlists */
 498     for(iidx=0; iidx<nri; iidx++)
 499     {
 500         /* Load shift vector for this list */
 501         i_shift_offset   = DIM*shiftidx[iidx];
 502
 503         /* Load limits for loop over neighbors */
 504         j_index_start    = jindex[iidx];
 505         j_index_end      = jindex[iidx+1];
 506
 507         /* Get outer coordinate index */
 508         inr              = iinr[iidx];
 509         i_coord_offset   = DIM*inr;
 510
 511         /* Load i particle coords and add shift vector */
 512         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 513
 514         fix0             = _mm_setzero_ps();
 515         fiy0             = _mm_setzero_ps();
 516         fiz0             = _mm_setzero_ps();
 517
 518         /* Load parameters for i particles */
 519         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 520
 521         /* Start inner kernel loop */
 522         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 523         {
 524
 525             /* Get j neighbor index, and coordinate index */
 526             jnrA             = jjnr[jidx];
 527             jnrB             = jjnr[jidx+1];
 528             jnrC             = jjnr[jidx+2];
 529             jnrD             = jjnr[jidx+3];
 530             j_coord_offsetA  = DIM*jnrA;
 531             j_coord_offsetB  = DIM*jnrB;
 532             j_coord_offsetC  = DIM*jnrC;
 533             j_coord_offsetD  = DIM*jnrD;
 534
 535             /* load j atom coordinates */
 536             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 537                                               x+j_coord_offsetC,x+j_coord_offsetD,
 538                                               &jx0,&jy0,&jz0);
 539
 540             /* Calculate displacement vector */
 541             dx00             = _mm_sub_ps(ix0,jx0);
 542             dy00             = _mm_sub_ps(iy0,jy0);
 543             dz00             = _mm_sub_ps(iz0,jz0);
 544
 545             /* Calculate squared distance and things based on it */
 546             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 547
 548             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 549
 550             /* Load parameters for j particles */
 551             vdwjidx0A        = 2*vdwtype[jnrA+0];
 552             vdwjidx0B        = 2*vdwtype[jnrB+0];
 553             vdwjidx0C        = 2*vdwtype[jnrC+0];
 554             vdwjidx0D        = 2*vdwtype[jnrD+0];
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r00              = _mm_mul_ps(rsq00,rinv00);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 564                                          vdwparam+vdwioffset0+vdwjidx0B,
 565                                          vdwparam+vdwioffset0+vdwjidx0C,
 566                                          vdwparam+vdwioffset0+vdwjidx0D,
 567                                          &c6_00,&c12_00);
 568
 569             /* Calculate table index by multiplying r with table scale and truncate to integer */
 570             rt               = _mm_mul_ps(r00,vftabscale);
 571             vfitab           = _mm_cvttps_epi32(rt);
 572             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 573             vfitab           = _mm_slli_epi32(vfitab,3);
 574
 575             /* CUBIC SPLINE TABLE DISPERSION */
 576             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 577             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 578             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 579             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 580             _MM_TRANSPOSE4_PS(Y,F,G,H);
 581             Heps             = _mm_mul_ps(vfeps,H);
 582             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 583             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 584             fvdw6            = _mm_mul_ps(c6_00,FF);
 585
 586             /* CUBIC SPLINE TABLE REPULSION */
 587             vfitab           = _mm_add_epi32(vfitab,ifour);
 588             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 589             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 590             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 591             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 592             _MM_TRANSPOSE4_PS(Y,F,G,H);
 593             Heps             = _mm_mul_ps(vfeps,H);
 594             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 595             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 596             fvdw12           = _mm_mul_ps(c12_00,FF);
 597             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 598
 599             fscal            = fvdw;
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_ps(fscal,dx00);
 603             ty               = _mm_mul_ps(fscal,dy00);
 604             tz               = _mm_mul_ps(fscal,dz00);
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_add_ps(fix0,tx);
 608             fiy0             = _mm_add_ps(fiy0,ty);
 609             fiz0             = _mm_add_ps(fiz0,tz);
 610
 611             fjptrA             = f+j_coord_offsetA;
 612             fjptrB             = f+j_coord_offsetB;
 613             fjptrC             = f+j_coord_offsetC;
 614             fjptrD             = f+j_coord_offsetD;
 615             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 616
 617             /* Inner loop uses 48 flops */
 618         }
 619
 620         if(jidx<j_index_end)
 621         {
 622
 623             /* Get j neighbor index, and coordinate index */
 624             jnrlistA         = jjnr[jidx];
 625             jnrlistB         = jjnr[jidx+1];
 626             jnrlistC         = jjnr[jidx+2];
 627             jnrlistD         = jjnr[jidx+3];
 628             /* Sign of each element will be negative for non-real atoms.
 629              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 630              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 631              */
 632             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 633             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 634             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 635             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 636             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 637             j_coord_offsetA  = DIM*jnrA;
 638             j_coord_offsetB  = DIM*jnrB;
 639             j_coord_offsetC  = DIM*jnrC;
 640             j_coord_offsetD  = DIM*jnrD;
 641
 642             /* load j atom coordinates */
 643             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 644                                               x+j_coord_offsetC,x+j_coord_offsetD,
 645                                               &jx0,&jy0,&jz0);
 646
 647             /* Calculate displacement vector */
 648             dx00             = _mm_sub_ps(ix0,jx0);
 649             dy00             = _mm_sub_ps(iy0,jy0);
 650             dz00             = _mm_sub_ps(iz0,jz0);
 651
 652             /* Calculate squared distance and things based on it */
 653             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 654
 655             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 656
 657             /* Load parameters for j particles */
 658             vdwjidx0A        = 2*vdwtype[jnrA+0];
 659             vdwjidx0B        = 2*vdwtype[jnrB+0];
 660             vdwjidx0C        = 2*vdwtype[jnrC+0];
 661             vdwjidx0D        = 2*vdwtype[jnrD+0];
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             r00              = _mm_mul_ps(rsq00,rinv00);
 668             r00              = _mm_andnot_ps(dummy_mask,r00);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 672                                          vdwparam+vdwioffset0+vdwjidx0B,
 673                                          vdwparam+vdwioffset0+vdwjidx0C,
 674                                          vdwparam+vdwioffset0+vdwjidx0D,
 675                                          &c6_00,&c12_00);
 676
 677             /* Calculate table index by multiplying r with table scale and truncate to integer */
 678             rt               = _mm_mul_ps(r00,vftabscale);
 679             vfitab           = _mm_cvttps_epi32(rt);
 680             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 681             vfitab           = _mm_slli_epi32(vfitab,3);
 682
 683             /* CUBIC SPLINE TABLE DISPERSION */
 684             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 685             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 686             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 687             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 688             _MM_TRANSPOSE4_PS(Y,F,G,H);
 689             Heps             = _mm_mul_ps(vfeps,H);
 690             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 691             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 692             fvdw6            = _mm_mul_ps(c6_00,FF);
 693
 694             /* CUBIC SPLINE TABLE REPULSION */
 695             vfitab           = _mm_add_epi32(vfitab,ifour);
 696             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 697             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 698             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 699             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 700             _MM_TRANSPOSE4_PS(Y,F,G,H);
 701             Heps             = _mm_mul_ps(vfeps,H);
 702             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 703             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 704             fvdw12           = _mm_mul_ps(c12_00,FF);
 705             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 706
 707             fscal            = 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 49 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 6 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_VDW_F,outeriter*6 + inneriter*49);
 747 }