src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse2_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_vdw->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->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             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 201                                                               charge+jnrC+0,charge+jnrD+0);
 202             vdwjidx0A        = 2*vdwtype[jnrA+0];
 203             vdwjidx0B        = 2*vdwtype[jnrB+0];
 204             vdwjidx0C        = 2*vdwtype[jnrC+0];
 205             vdwjidx0D        = 2*vdwtype[jnrD+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             r00              = _mm_mul_ps(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_ps(iq0,jq0);
 215             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 216                                          vdwparam+vdwioffset0+vdwjidx0B,
 217                                          vdwparam+vdwioffset0+vdwjidx0C,
 218                                          vdwparam+vdwioffset0+vdwjidx0D,
 219                                          &c6_00,&c12_00);
 220
 221             /* Calculate table index by multiplying r with table scale and truncate to integer */
 222             rt               = _mm_mul_ps(r00,vftabscale);
 223             vfitab           = _mm_cvttps_epi32(rt);
 224             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 225             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 226
 227             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 228             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 229             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 230             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 231             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 232             _MM_TRANSPOSE4_PS(Y,F,G,H);
 233             Heps             = _mm_mul_ps(vfeps,H);
 234             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 235             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 236             velec            = _mm_mul_ps(qq00,VV);
 237             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 238             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 239
 240             /* CUBIC SPLINE TABLE DISPERSION */
 241             vfitab           = _mm_add_epi32(vfitab,ifour);
 242             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 243             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 244             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 245             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 246             _MM_TRANSPOSE4_PS(Y,F,G,H);
 247             Heps             = _mm_mul_ps(vfeps,H);
 248             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 249             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 250             vvdw6            = _mm_mul_ps(c6_00,VV);
 251             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 252             fvdw6            = _mm_mul_ps(c6_00,FF);
 253
 254             /* CUBIC SPLINE TABLE REPULSION */
 255             vfitab           = _mm_add_epi32(vfitab,ifour);
 256             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 259             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 260             _MM_TRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm_mul_ps(vfeps,H);
 262             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 263             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 264             vvdw12           = _mm_mul_ps(c12_00,VV);
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             fvdw12           = _mm_mul_ps(c12_00,FF);
 267             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 268             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velecsum         = _mm_add_ps(velecsum,velec);
 272             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 273
 274             fscal            = _mm_add_ps(felec,fvdw);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_ps(fscal,dx00);
 278             ty               = _mm_mul_ps(fscal,dy00);
 279             tz               = _mm_mul_ps(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_add_ps(fix0,tx);
 283             fiy0             = _mm_add_ps(fiy0,ty);
 284             fiz0             = _mm_add_ps(fiz0,tz);
 285
 286             fjptrA             = f+j_coord_offsetA;
 287             fjptrB             = f+j_coord_offsetB;
 288             fjptrC             = f+j_coord_offsetC;
 289             fjptrD             = f+j_coord_offsetD;
 290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 291
 292             /* Inner loop uses 73 flops */
 293         }
 294
 295         if(jidx<j_index_end)
 296         {
 297
 298             /* Get j neighbor index, and coordinate index */
 299             jnrlistA         = jjnr[jidx];
 300             jnrlistB         = jjnr[jidx+1];
 301             jnrlistC         = jjnr[jidx+2];
 302             jnrlistD         = jjnr[jidx+3];
 303             /* Sign of each element will be negative for non-real atoms.
 304              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 305              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 306              */
 307             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 308             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 309             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 310             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 311             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 312             j_coord_offsetA  = DIM*jnrA;
 313             j_coord_offsetB  = DIM*jnrB;
 314             j_coord_offsetC  = DIM*jnrC;
 315             j_coord_offsetD  = DIM*jnrD;
 316
 317             /* load j atom coordinates */
 318             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 319                                               x+j_coord_offsetC,x+j_coord_offsetD,
 320                                               &jx0,&jy0,&jz0);
 321
 322             /* Calculate displacement vector */
 323             dx00             = _mm_sub_ps(ix0,jx0);
 324             dy00             = _mm_sub_ps(iy0,jy0);
 325             dz00             = _mm_sub_ps(iz0,jz0);
 326
 327             /* Calculate squared distance and things based on it */
 328             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 329
 330             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 331
 332             /* Load parameters for j particles */
 333             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 334                                                               charge+jnrC+0,charge+jnrD+0);
 335             vdwjidx0A        = 2*vdwtype[jnrA+0];
 336             vdwjidx0B        = 2*vdwtype[jnrB+0];
 337             vdwjidx0C        = 2*vdwtype[jnrC+0];
 338             vdwjidx0D        = 2*vdwtype[jnrD+0];
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             r00              = _mm_mul_ps(rsq00,rinv00);
 345             r00              = _mm_andnot_ps(dummy_mask,r00);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq00             = _mm_mul_ps(iq0,jq0);
 349             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 350                                          vdwparam+vdwioffset0+vdwjidx0B,
 351                                          vdwparam+vdwioffset0+vdwjidx0C,
 352                                          vdwparam+vdwioffset0+vdwjidx0D,
 353                                          &c6_00,&c12_00);
 354
 355             /* Calculate table index by multiplying r with table scale and truncate to integer */
 356             rt               = _mm_mul_ps(r00,vftabscale);
 357             vfitab           = _mm_cvttps_epi32(rt);
 358             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 359             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 360
 361             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 362             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 363             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 364             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 365             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 366             _MM_TRANSPOSE4_PS(Y,F,G,H);
 367             Heps             = _mm_mul_ps(vfeps,H);
 368             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 369             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 370             velec            = _mm_mul_ps(qq00,VV);
 371             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 372             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 373
 374             /* CUBIC SPLINE TABLE DISPERSION */
 375             vfitab           = _mm_add_epi32(vfitab,ifour);
 376             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 377             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 378             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 379             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 380             _MM_TRANSPOSE4_PS(Y,F,G,H);
 381             Heps             = _mm_mul_ps(vfeps,H);
 382             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 383             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 384             vvdw6            = _mm_mul_ps(c6_00,VV);
 385             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 386             fvdw6            = _mm_mul_ps(c6_00,FF);
 387
 388             /* CUBIC SPLINE TABLE REPULSION */
 389             vfitab           = _mm_add_epi32(vfitab,ifour);
 390             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 391             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 392             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 393             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 394             _MM_TRANSPOSE4_PS(Y,F,G,H);
 395             Heps             = _mm_mul_ps(vfeps,H);
 396             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 397             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 398             vvdw12           = _mm_mul_ps(c12_00,VV);
 399             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 400             fvdw12           = _mm_mul_ps(c12_00,FF);
 401             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 402             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_andnot_ps(dummy_mask,velec);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 408             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 409
 410             fscal            = _mm_add_ps(felec,fvdw);
 411
 412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_ps(fscal,dx00);
 416             ty               = _mm_mul_ps(fscal,dy00);
 417             tz               = _mm_mul_ps(fscal,dz00);
 418
 419             /* Update vectorial force */
 420             fix0             = _mm_add_ps(fix0,tx);
 421             fiy0             = _mm_add_ps(fiy0,ty);
 422             fiz0             = _mm_add_ps(fiz0,tz);
 423
 424             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 425             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 426             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 427             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 429
 430             /* Inner loop uses 74 flops */
 431         }
 432
 433         /* End of innermost loop */
 434
 435         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 436                                               f+i_coord_offset,fshift+i_shift_offset);
 437
 438         ggid                        = gid[iidx];
 439         /* Update potential energies */
 440         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 441         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 442
 443         /* Increment number of inner iterations */
 444         inneriter                  += j_index_end - j_index_start;
 445
 446         /* Outer loop uses 9 flops */
 447     }
 448
 449     /* Increment number of outer iterations */
 450     outeriter        += nri;
 451
 452     /* Update outer/inner flops */
 453
 454     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 455 }
 456 /*
 457  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 458  * Electrostatics interaction: CubicSplineTable
 459  * VdW interaction:            CubicSplineTable
 460  * Geometry:                   Particle-Particle
 461  * Calculate force/pot:        Force
 462  */
 463 void
 464 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 465                     (t_nblist                    * gmx_restrict       nlist,
 466                      rvec                        * gmx_restrict          xx,
 467                      rvec                        * gmx_restrict          ff,
 468                      t_forcerec                  * gmx_restrict          fr,
 469                      t_mdatoms                   * gmx_restrict     mdatoms,
 470                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 471                      t_nrnb                      * gmx_restrict        nrnb)
 472 {
 473     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 474      * just 0 for non-waters.
 475      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 476      * jnr indices corresponding to data put in the four positions in the SIMD register.
 477      */
 478     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 479     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 480     int              jnrA,jnrB,jnrC,jnrD;
 481     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 482     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 483     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 484     real             rcutoff_scalar;
 485     real             *shiftvec,*fshift,*x,*f;
 486     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 487     real             scratch[4*DIM];
 488     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 489     int              vdwioffset0;
 490     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 491     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 492     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 493     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 494     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 495     real             *charge;
 496     int              nvdwtype;
 497     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 498     int              *vdwtype;
 499     real             *vdwparam;
 500     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 501     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 502     __m128i          vfitab;
 503     __m128i          ifour       = _mm_set1_epi32(4);
 504     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 505     real             *vftab;
 506     __m128           dummy_mask,cutoff_mask;
 507     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 508     __m128           one     = _mm_set1_ps(1.0);
 509     __m128           two     = _mm_set1_ps(2.0);
 510     x                = xx[0];
 511     f                = ff[0];
 512
 513     nri              = nlist->nri;
 514     iinr             = nlist->iinr;
 515     jindex           = nlist->jindex;
 516     jjnr             = nlist->jjnr;
 517     shiftidx         = nlist->shift;
 518     gid              = nlist->gid;
 519     shiftvec         = fr->shift_vec[0];
 520     fshift           = fr->fshift[0];
 521     facel            = _mm_set1_ps(fr->epsfac);
 522     charge           = mdatoms->chargeA;
 523     nvdwtype         = fr->ntype;
 524     vdwparam         = fr->nbfp;
 525     vdwtype          = mdatoms->typeA;
 526
 527     vftab            = kernel_data->table_elec_vdw->data;
 528     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 529
 530     /* Avoid stupid compiler warnings */
 531     jnrA = jnrB = jnrC = jnrD = 0;
 532     j_coord_offsetA = 0;
 533     j_coord_offsetB = 0;
 534     j_coord_offsetC = 0;
 535     j_coord_offsetD = 0;
 536
 537     outeriter        = 0;
 538     inneriter        = 0;
 539
 540     for(iidx=0;iidx<4*DIM;iidx++)
 541     {
 542         scratch[iidx] = 0.0;
 543     }
 544
 545     /* Start outer loop over neighborlists */
 546     for(iidx=0; iidx<nri; iidx++)
 547     {
 548         /* Load shift vector for this list */
 549         i_shift_offset   = DIM*shiftidx[iidx];
 550
 551         /* Load limits for loop over neighbors */
 552         j_index_start    = jindex[iidx];
 553         j_index_end      = jindex[iidx+1];
 554
 555         /* Get outer coordinate index */
 556         inr              = iinr[iidx];
 557         i_coord_offset   = DIM*inr;
 558
 559         /* Load i particle coords and add shift vector */
 560         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 561
 562         fix0             = _mm_setzero_ps();
 563         fiy0             = _mm_setzero_ps();
 564         fiz0             = _mm_setzero_ps();
 565
 566         /* Load parameters for i particles */
 567         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 568         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 569
 570         /* Start inner kernel loop */
 571         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 572         {
 573
 574             /* Get j neighbor index, and coordinate index */
 575             jnrA             = jjnr[jidx];
 576             jnrB             = jjnr[jidx+1];
 577             jnrC             = jjnr[jidx+2];
 578             jnrD             = jjnr[jidx+3];
 579             j_coord_offsetA  = DIM*jnrA;
 580             j_coord_offsetB  = DIM*jnrB;
 581             j_coord_offsetC  = DIM*jnrC;
 582             j_coord_offsetD  = DIM*jnrD;
 583
 584             /* load j atom coordinates */
 585             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 586                                               x+j_coord_offsetC,x+j_coord_offsetD,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _mm_sub_ps(ix0,jx0);
 591             dy00             = _mm_sub_ps(iy0,jy0);
 592             dz00             = _mm_sub_ps(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 601                                                               charge+jnrC+0,charge+jnrD+0);
 602             vdwjidx0A        = 2*vdwtype[jnrA+0];
 603             vdwjidx0B        = 2*vdwtype[jnrB+0];
 604             vdwjidx0C        = 2*vdwtype[jnrC+0];
 605             vdwjidx0D        = 2*vdwtype[jnrD+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r00              = _mm_mul_ps(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _mm_mul_ps(iq0,jq0);
 615             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,
 617                                          vdwparam+vdwioffset0+vdwjidx0C,
 618                                          vdwparam+vdwioffset0+vdwjidx0D,
 619                                          &c6_00,&c12_00);
 620
 621             /* Calculate table index by multiplying r with table scale and truncate to integer */
 622             rt               = _mm_mul_ps(r00,vftabscale);
 623             vfitab           = _mm_cvttps_epi32(rt);
 624             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 625             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 626
 627             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 628             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 629             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 630             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 631             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 632             _MM_TRANSPOSE4_PS(Y,F,G,H);
 633             Heps             = _mm_mul_ps(vfeps,H);
 634             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 635             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 636             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 637
 638             /* CUBIC SPLINE TABLE DISPERSION */
 639             vfitab           = _mm_add_epi32(vfitab,ifour);
 640             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 641             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 642             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 643             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 644             _MM_TRANSPOSE4_PS(Y,F,G,H);
 645             Heps             = _mm_mul_ps(vfeps,H);
 646             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 647             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 648             fvdw6            = _mm_mul_ps(c6_00,FF);
 649
 650             /* CUBIC SPLINE TABLE REPULSION */
 651             vfitab           = _mm_add_epi32(vfitab,ifour);
 652             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 653             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 654             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 655             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 656             _MM_TRANSPOSE4_PS(Y,F,G,H);
 657             Heps             = _mm_mul_ps(vfeps,H);
 658             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 659             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 660             fvdw12           = _mm_mul_ps(c12_00,FF);
 661             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 662
 663             fscal            = _mm_add_ps(felec,fvdw);
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_ps(fscal,dx00);
 667             ty               = _mm_mul_ps(fscal,dy00);
 668             tz               = _mm_mul_ps(fscal,dz00);
 669
 670             /* Update vectorial force */
 671             fix0             = _mm_add_ps(fix0,tx);
 672             fiy0             = _mm_add_ps(fiy0,ty);
 673             fiz0             = _mm_add_ps(fiz0,tz);
 674
 675             fjptrA             = f+j_coord_offsetA;
 676             fjptrB             = f+j_coord_offsetB;
 677             fjptrC             = f+j_coord_offsetC;
 678             fjptrD             = f+j_coord_offsetD;
 679             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 680
 681             /* Inner loop uses 61 flops */
 682         }
 683
 684         if(jidx<j_index_end)
 685         {
 686
 687             /* Get j neighbor index, and coordinate index */
 688             jnrlistA         = jjnr[jidx];
 689             jnrlistB         = jjnr[jidx+1];
 690             jnrlistC         = jjnr[jidx+2];
 691             jnrlistD         = jjnr[jidx+3];
 692             /* Sign of each element will be negative for non-real atoms.
 693              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 694              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 695              */
 696             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 697             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 698             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 699             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 700             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 701             j_coord_offsetA  = DIM*jnrA;
 702             j_coord_offsetB  = DIM*jnrB;
 703             j_coord_offsetC  = DIM*jnrC;
 704             j_coord_offsetD  = DIM*jnrD;
 705
 706             /* load j atom coordinates */
 707             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 708                                               x+j_coord_offsetC,x+j_coord_offsetD,
 709                                               &jx0,&jy0,&jz0);
 710
 711             /* Calculate displacement vector */
 712             dx00             = _mm_sub_ps(ix0,jx0);
 713             dy00             = _mm_sub_ps(iy0,jy0);
 714             dz00             = _mm_sub_ps(iz0,jz0);
 715
 716             /* Calculate squared distance and things based on it */
 717             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 718
 719             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 720
 721             /* Load parameters for j particles */
 722             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 723                                                               charge+jnrC+0,charge+jnrD+0);
 724             vdwjidx0A        = 2*vdwtype[jnrA+0];
 725             vdwjidx0B        = 2*vdwtype[jnrB+0];
 726             vdwjidx0C        = 2*vdwtype[jnrC+0];
 727             vdwjidx0D        = 2*vdwtype[jnrD+0];
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             r00              = _mm_mul_ps(rsq00,rinv00);
 734             r00              = _mm_andnot_ps(dummy_mask,r00);
 735
 736             /* Compute parameters for interactions between i and j atoms */
 737             qq00             = _mm_mul_ps(iq0,jq0);
 738             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 739                                          vdwparam+vdwioffset0+vdwjidx0B,
 740                                          vdwparam+vdwioffset0+vdwjidx0C,
 741                                          vdwparam+vdwioffset0+vdwjidx0D,
 742                                          &c6_00,&c12_00);
 743
 744             /* Calculate table index by multiplying r with table scale and truncate to integer */
 745             rt               = _mm_mul_ps(r00,vftabscale);
 746             vfitab           = _mm_cvttps_epi32(rt);
 747             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 748             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 749
 750             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 751             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 752             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 753             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 754             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 755             _MM_TRANSPOSE4_PS(Y,F,G,H);
 756             Heps             = _mm_mul_ps(vfeps,H);
 757             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 758             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 759             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 760
 761             /* CUBIC SPLINE TABLE DISPERSION */
 762             vfitab           = _mm_add_epi32(vfitab,ifour);
 763             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 764             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 765             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 766             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 767             _MM_TRANSPOSE4_PS(Y,F,G,H);
 768             Heps             = _mm_mul_ps(vfeps,H);
 769             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 770             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 771             fvdw6            = _mm_mul_ps(c6_00,FF);
 772
 773             /* CUBIC SPLINE TABLE REPULSION */
 774             vfitab           = _mm_add_epi32(vfitab,ifour);
 775             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 776             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 777             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 778             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 779             _MM_TRANSPOSE4_PS(Y,F,G,H);
 780             Heps             = _mm_mul_ps(vfeps,H);
 781             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 782             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 783             fvdw12           = _mm_mul_ps(c12_00,FF);
 784             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 785
 786             fscal            = _mm_add_ps(felec,fvdw);
 787
 788             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 789
 790             /* Calculate temporary vectorial force */
 791             tx               = _mm_mul_ps(fscal,dx00);
 792             ty               = _mm_mul_ps(fscal,dy00);
 793             tz               = _mm_mul_ps(fscal,dz00);
 794
 795             /* Update vectorial force */
 796             fix0             = _mm_add_ps(fix0,tx);
 797             fiy0             = _mm_add_ps(fiy0,ty);
 798             fiz0             = _mm_add_ps(fiz0,tz);
 799
 800             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 801             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 802             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 803             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 804             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 805
 806             /* Inner loop uses 62 flops */
 807         }
 808
 809         /* End of innermost loop */
 810
 811         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 812                                               f+i_coord_offset,fshift+i_shift_offset);
 813
 814         /* Increment number of inner iterations */
 815         inneriter                  += j_index_end - j_index_start;
 816
 817         /* Outer loop uses 7 flops */
 818     }
 819
 820     /* Increment number of outer iterations */
 821     outeriter        += nri;
 822
 823     /* Update outer/inner flops */
 824
 825     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 826 }