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) 2012,2013,2014,2015,2017, 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 sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: None
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  92     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128           dummy_mask,cutoff_mask;
  98     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  99     __m128           one     = _mm_set1_ps(1.0);
 100     __m128           two     = _mm_set1_ps(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_ps();
 152         fiy0             = _mm_setzero_ps();
 153         fiz0             = _mm_setzero_ps();
 154
 155         /* Load parameters for i particles */
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         vvdwsum          = _mm_setzero_ps();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             jnrC             = jjnr[jidx+2];
 169             jnrD             = jjnr[jidx+3];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               x+j_coord_offsetC,x+j_coord_offsetD,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_ps(ix0,jx0);
 182             dy00             = _mm_sub_ps(iy0,jy0);
 183             dz00             = _mm_sub_ps(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 187
 188             rinv00           = sse41_invsqrt_f(rsq00);
 189
 190             /* Load parameters for j particles */
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193             vdwjidx0C        = 2*vdwtype[jnrC+0];
 194             vdwjidx0D        = 2*vdwtype[jnrD+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             r00              = _mm_mul_ps(rsq00,rinv00);
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 204                                          vdwparam+vdwioffset0+vdwjidx0B,
 205                                          vdwparam+vdwioffset0+vdwjidx0C,
 206                                          vdwparam+vdwioffset0+vdwjidx0D,
 207                                          &c6_00,&c12_00);
 208
 209             /* Calculate table index by multiplying r with table scale and truncate to integer */
 210             rt               = _mm_mul_ps(r00,vftabscale);
 211             vfitab           = _mm_cvttps_epi32(rt);
 212             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 213             vfitab           = _mm_slli_epi32(vfitab,3);
 214
 215             /* CUBIC SPLINE TABLE DISPERSION */
 216             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 217             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 218             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 219             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 220             _MM_TRANSPOSE4_PS(Y,F,G,H);
 221             Heps             = _mm_mul_ps(vfeps,H);
 222             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 223             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 224             vvdw6            = _mm_mul_ps(c6_00,VV);
 225             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 226             fvdw6            = _mm_mul_ps(c6_00,FF);
 227
 228             /* CUBIC SPLINE TABLE REPULSION */
 229             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_ps(c12_00,VV);
 239             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 240             fvdw12           = _mm_mul_ps(c12_00,FF);
 241             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 242             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 246
 247             fscal            = fvdw;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_ps(fscal,dx00);
 251             ty               = _mm_mul_ps(fscal,dy00);
 252             tz               = _mm_mul_ps(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_ps(fix0,tx);
 256             fiy0             = _mm_add_ps(fiy0,ty);
 257             fiz0             = _mm_add_ps(fiz0,tz);
 258
 259             fjptrA             = f+j_coord_offsetA;
 260             fjptrB             = f+j_coord_offsetB;
 261             fjptrC             = f+j_coord_offsetC;
 262             fjptrD             = f+j_coord_offsetD;
 263             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 264
 265             /* Inner loop uses 56 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             /* Get j neighbor index, and coordinate index */
 272             jnrlistA         = jjnr[jidx];
 273             jnrlistB         = jjnr[jidx+1];
 274             jnrlistC         = jjnr[jidx+2];
 275             jnrlistD         = jjnr[jidx+3];
 276             /* Sign of each element will be negative for non-real atoms.
 277              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 278              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 279              */
 280             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 281             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 282             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 283             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 284             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 285             j_coord_offsetA  = DIM*jnrA;
 286             j_coord_offsetB  = DIM*jnrB;
 287             j_coord_offsetC  = DIM*jnrC;
 288             j_coord_offsetD  = DIM*jnrD;
 289
 290             /* load j atom coordinates */
 291             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 292                                               x+j_coord_offsetC,x+j_coord_offsetD,
 293                                               &jx0,&jy0,&jz0);
 294
 295             /* Calculate displacement vector */
 296             dx00             = _mm_sub_ps(ix0,jx0);
 297             dy00             = _mm_sub_ps(iy0,jy0);
 298             dz00             = _mm_sub_ps(iz0,jz0);
 299
 300             /* Calculate squared distance and things based on it */
 301             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 302
 303             rinv00           = sse41_invsqrt_f(rsq00);
 304
 305             /* Load parameters for j particles */
 306             vdwjidx0A        = 2*vdwtype[jnrA+0];
 307             vdwjidx0B        = 2*vdwtype[jnrB+0];
 308             vdwjidx0C        = 2*vdwtype[jnrC+0];
 309             vdwjidx0D        = 2*vdwtype[jnrD+0];
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r00              = _mm_mul_ps(rsq00,rinv00);
 316             r00              = _mm_andnot_ps(dummy_mask,r00);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 320                                          vdwparam+vdwioffset0+vdwjidx0B,
 321                                          vdwparam+vdwioffset0+vdwjidx0C,
 322                                          vdwparam+vdwioffset0+vdwjidx0D,
 323                                          &c6_00,&c12_00);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm_mul_ps(r00,vftabscale);
 327             vfitab           = _mm_cvttps_epi32(rt);
 328             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 329             vfitab           = _mm_slli_epi32(vfitab,3);
 330
 331             /* CUBIC SPLINE TABLE DISPERSION */
 332             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 333             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 334             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 335             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 336             _MM_TRANSPOSE4_PS(Y,F,G,H);
 337             Heps             = _mm_mul_ps(vfeps,H);
 338             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 339             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 340             vvdw6            = _mm_mul_ps(c6_00,VV);
 341             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 342             fvdw6            = _mm_mul_ps(c6_00,FF);
 343
 344             /* CUBIC SPLINE TABLE REPULSION */
 345             vfitab           = _mm_add_epi32(vfitab,ifour);
 346             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 347             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 348             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 349             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 350             _MM_TRANSPOSE4_PS(Y,F,G,H);
 351             Heps             = _mm_mul_ps(vfeps,H);
 352             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 353             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 354             vvdw12           = _mm_mul_ps(c12_00,VV);
 355             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 356             fvdw12           = _mm_mul_ps(c12_00,FF);
 357             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 358             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 362             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 363
 364             fscal            = fvdw;
 365
 366             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm_mul_ps(fscal,dx00);
 370             ty               = _mm_mul_ps(fscal,dy00);
 371             tz               = _mm_mul_ps(fscal,dz00);
 372
 373             /* Update vectorial force */
 374             fix0             = _mm_add_ps(fix0,tx);
 375             fiy0             = _mm_add_ps(fiy0,ty);
 376             fiz0             = _mm_add_ps(fiz0,tz);
 377
 378             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 379             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 380             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 381             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 382             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 383
 384             /* Inner loop uses 57 flops */
 385         }
 386
 387         /* End of innermost loop */
 388
 389         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 390                                               f+i_coord_offset,fshift+i_shift_offset);
 391
 392         ggid                        = gid[iidx];
 393         /* Update potential energies */
 394         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 395
 396         /* Increment number of inner iterations */
 397         inneriter                  += j_index_end - j_index_start;
 398
 399         /* Outer loop uses 7 flops */
 400     }
 401
 402     /* Increment number of outer iterations */
 403     outeriter        += nri;
 404
 405     /* Update outer/inner flops */
 406
 407     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*57);
 408 }
 409 /*
 410  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_single
 411  * Electrostatics interaction: None
 412  * VdW interaction:            CubicSplineTable
 413  * Geometry:                   Particle-Particle
 414  * Calculate force/pot:        Force
 415  */
 416 void
 417 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_single
 418                     (t_nblist                    * gmx_restrict       nlist,
 419                      rvec                        * gmx_restrict          xx,
 420                      rvec                        * gmx_restrict          ff,
 421                      struct t_forcerec           * gmx_restrict          fr,
 422                      t_mdatoms                   * gmx_restrict     mdatoms,
 423                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 424                      t_nrnb                      * gmx_restrict        nrnb)
 425 {
 426     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 427      * just 0 for non-waters.
 428      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 429      * jnr indices corresponding to data put in the four positions in the SIMD register.
 430      */
 431     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 432     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 433     int              jnrA,jnrB,jnrC,jnrD;
 434     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 435     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 436     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 437     real             rcutoff_scalar;
 438     real             *shiftvec,*fshift,*x,*f;
 439     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 440     real             scratch[4*DIM];
 441     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 442     int              vdwioffset0;
 443     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 444     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 445     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 446     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 447     int              nvdwtype;
 448     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 449     int              *vdwtype;
 450     real             *vdwparam;
 451     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 452     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 453     __m128i          vfitab;
 454     __m128i          ifour       = _mm_set1_epi32(4);
 455     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 456     real             *vftab;
 457     __m128           dummy_mask,cutoff_mask;
 458     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 459     __m128           one     = _mm_set1_ps(1.0);
 460     __m128           two     = _mm_set1_ps(2.0);
 461     x                = xx[0];
 462     f                = ff[0];
 463
 464     nri              = nlist->nri;
 465     iinr             = nlist->iinr;
 466     jindex           = nlist->jindex;
 467     jjnr             = nlist->jjnr;
 468     shiftidx         = nlist->shift;
 469     gid              = nlist->gid;
 470     shiftvec         = fr->shift_vec[0];
 471     fshift           = fr->fshift[0];
 472     nvdwtype         = fr->ntype;
 473     vdwparam         = fr->nbfp;
 474     vdwtype          = mdatoms->typeA;
 475
 476     vftab            = kernel_data->table_vdw->data;
 477     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 478
 479     /* Avoid stupid compiler warnings */
 480     jnrA = jnrB = jnrC = jnrD = 0;
 481     j_coord_offsetA = 0;
 482     j_coord_offsetB = 0;
 483     j_coord_offsetC = 0;
 484     j_coord_offsetD = 0;
 485
 486     outeriter        = 0;
 487     inneriter        = 0;
 488
 489     for(iidx=0;iidx<4*DIM;iidx++)
 490     {
 491         scratch[iidx] = 0.0;
 492     }
 493
 494     /* Start outer loop over neighborlists */
 495     for(iidx=0; iidx<nri; iidx++)
 496     {
 497         /* Load shift vector for this list */
 498         i_shift_offset   = DIM*shiftidx[iidx];
 499
 500         /* Load limits for loop over neighbors */
 501         j_index_start    = jindex[iidx];
 502         j_index_end      = jindex[iidx+1];
 503
 504         /* Get outer coordinate index */
 505         inr              = iinr[iidx];
 506         i_coord_offset   = DIM*inr;
 507
 508         /* Load i particle coords and add shift vector */
 509         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 510
 511         fix0             = _mm_setzero_ps();
 512         fiy0             = _mm_setzero_ps();
 513         fiz0             = _mm_setzero_ps();
 514
 515         /* Load parameters for i particles */
 516         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 517
 518         /* Start inner kernel loop */
 519         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 520         {
 521
 522             /* Get j neighbor index, and coordinate index */
 523             jnrA             = jjnr[jidx];
 524             jnrB             = jjnr[jidx+1];
 525             jnrC             = jjnr[jidx+2];
 526             jnrD             = jjnr[jidx+3];
 527             j_coord_offsetA  = DIM*jnrA;
 528             j_coord_offsetB  = DIM*jnrB;
 529             j_coord_offsetC  = DIM*jnrC;
 530             j_coord_offsetD  = DIM*jnrD;
 531
 532             /* load j atom coordinates */
 533             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                               x+j_coord_offsetC,x+j_coord_offsetD,
 535                                               &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm_sub_ps(ix0,jx0);
 539             dy00             = _mm_sub_ps(iy0,jy0);
 540             dz00             = _mm_sub_ps(iz0,jz0);
 541
 542             /* Calculate squared distance and things based on it */
 543             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 544
 545             rinv00           = sse41_invsqrt_f(rsq00);
 546
 547             /* Load parameters for j particles */
 548             vdwjidx0A        = 2*vdwtype[jnrA+0];
 549             vdwjidx0B        = 2*vdwtype[jnrB+0];
 550             vdwjidx0C        = 2*vdwtype[jnrC+0];
 551             vdwjidx0D        = 2*vdwtype[jnrD+0];
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r00              = _mm_mul_ps(rsq00,rinv00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,
 562                                          vdwparam+vdwioffset0+vdwjidx0C,
 563                                          vdwparam+vdwioffset0+vdwjidx0D,
 564                                          &c6_00,&c12_00);
 565
 566             /* Calculate table index by multiplying r with table scale and truncate to integer */
 567             rt               = _mm_mul_ps(r00,vftabscale);
 568             vfitab           = _mm_cvttps_epi32(rt);
 569             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 570             vfitab           = _mm_slli_epi32(vfitab,3);
 571
 572             /* CUBIC SPLINE TABLE DISPERSION */
 573             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 574             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 575             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 576             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 577             _MM_TRANSPOSE4_PS(Y,F,G,H);
 578             Heps             = _mm_mul_ps(vfeps,H);
 579             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 580             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 581             fvdw6            = _mm_mul_ps(c6_00,FF);
 582
 583             /* CUBIC SPLINE TABLE REPULSION */
 584             vfitab           = _mm_add_epi32(vfitab,ifour);
 585             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 586             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 587             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 588             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 589             _MM_TRANSPOSE4_PS(Y,F,G,H);
 590             Heps             = _mm_mul_ps(vfeps,H);
 591             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 592             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 593             fvdw12           = _mm_mul_ps(c12_00,FF);
 594             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 595
 596             fscal            = fvdw;
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_ps(fscal,dx00);
 600             ty               = _mm_mul_ps(fscal,dy00);
 601             tz               = _mm_mul_ps(fscal,dz00);
 602
 603             /* Update vectorial force */
 604             fix0             = _mm_add_ps(fix0,tx);
 605             fiy0             = _mm_add_ps(fiy0,ty);
 606             fiz0             = _mm_add_ps(fiz0,tz);
 607
 608             fjptrA             = f+j_coord_offsetA;
 609             fjptrB             = f+j_coord_offsetB;
 610             fjptrC             = f+j_coord_offsetC;
 611             fjptrD             = f+j_coord_offsetD;
 612             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 613
 614             /* Inner loop uses 48 flops */
 615         }
 616
 617         if(jidx<j_index_end)
 618         {
 619
 620             /* Get j neighbor index, and coordinate index */
 621             jnrlistA         = jjnr[jidx];
 622             jnrlistB         = jjnr[jidx+1];
 623             jnrlistC         = jjnr[jidx+2];
 624             jnrlistD         = jjnr[jidx+3];
 625             /* Sign of each element will be negative for non-real atoms.
 626              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 627              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 628              */
 629             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 630             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 631             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 632             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 633             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 634             j_coord_offsetA  = DIM*jnrA;
 635             j_coord_offsetB  = DIM*jnrB;
 636             j_coord_offsetC  = DIM*jnrC;
 637             j_coord_offsetD  = DIM*jnrD;
 638
 639             /* load j atom coordinates */
 640             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 641                                               x+j_coord_offsetC,x+j_coord_offsetD,
 642                                               &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm_sub_ps(ix0,jx0);
 646             dy00             = _mm_sub_ps(iy0,jy0);
 647             dz00             = _mm_sub_ps(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 651
 652             rinv00           = sse41_invsqrt_f(rsq00);
 653
 654             /* Load parameters for j particles */
 655             vdwjidx0A        = 2*vdwtype[jnrA+0];
 656             vdwjidx0B        = 2*vdwtype[jnrB+0];
 657             vdwjidx0C        = 2*vdwtype[jnrC+0];
 658             vdwjidx0D        = 2*vdwtype[jnrD+0];
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             r00              = _mm_mul_ps(rsq00,rinv00);
 665             r00              = _mm_andnot_ps(dummy_mask,r00);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 669                                          vdwparam+vdwioffset0+vdwjidx0B,
 670                                          vdwparam+vdwioffset0+vdwjidx0C,
 671                                          vdwparam+vdwioffset0+vdwjidx0D,
 672                                          &c6_00,&c12_00);
 673
 674             /* Calculate table index by multiplying r with table scale and truncate to integer */
 675             rt               = _mm_mul_ps(r00,vftabscale);
 676             vfitab           = _mm_cvttps_epi32(rt);
 677             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 678             vfitab           = _mm_slli_epi32(vfitab,3);
 679
 680             /* CUBIC SPLINE TABLE DISPERSION */
 681             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 682             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 683             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 684             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 685             _MM_TRANSPOSE4_PS(Y,F,G,H);
 686             Heps             = _mm_mul_ps(vfeps,H);
 687             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 688             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 689             fvdw6            = _mm_mul_ps(c6_00,FF);
 690
 691             /* CUBIC SPLINE TABLE REPULSION */
 692             vfitab           = _mm_add_epi32(vfitab,ifour);
 693             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 694             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 695             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 696             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 697             _MM_TRANSPOSE4_PS(Y,F,G,H);
 698             Heps             = _mm_mul_ps(vfeps,H);
 699             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 700             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 701             fvdw12           = _mm_mul_ps(c12_00,FF);
 702             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 703
 704             fscal            = fvdw;
 705
 706             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 707
 708             /* Calculate temporary vectorial force */
 709             tx               = _mm_mul_ps(fscal,dx00);
 710             ty               = _mm_mul_ps(fscal,dy00);
 711             tz               = _mm_mul_ps(fscal,dz00);
 712
 713             /* Update vectorial force */
 714             fix0             = _mm_add_ps(fix0,tx);
 715             fiy0             = _mm_add_ps(fiy0,ty);
 716             fiz0             = _mm_add_ps(fiz0,tz);
 717
 718             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 719             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 720             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 721             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 722             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 723
 724             /* Inner loop uses 49 flops */
 725         }
 726
 727         /* End of innermost loop */
 728
 729         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 730                                               f+i_coord_offset,fshift+i_shift_offset);
 731
 732         /* Increment number of inner iterations */
 733         inneriter                  += j_index_end - j_index_start;
 734
 735         /* Outer loop uses 6 flops */
 736     }
 737
 738     /* Increment number of outer iterations */
 739     outeriter        += nri;
 740
 741     /* Update outer/inner flops */
 742
 743     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*49);
 744 }