src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          gbitab;
  74     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  75     __m128d          minushalf = _mm_set1_pd(-0.5);
  76     real             *invsqrta,*dvda,*gbtab;
  77     int              nvdwtype;
  78     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  82     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  83     __m128i          vfitab;
  84     __m128i          ifour       = _mm_set1_epi32(4);
  85     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  86     real             *vftab;
  87     __m128d          dummy_mask,cutoff_mask;
  88     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  89     __m128d          one     = _mm_set1_pd(1.0);
  90     __m128d          two     = _mm_set1_pd(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_pd(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104     nvdwtype         = fr->ntype;
 105     vdwparam         = fr->nbfp;
 106     vdwtype          = mdatoms->typeA;
 107
 108     vftab            = kernel_data->table_vdw->data;
 109     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 110
 111     invsqrta         = fr->invsqrta;
 112     dvda             = fr->dvda;
 113     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 114     gbtab            = fr->gbtab.data;
 115     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 148         isai0            = _mm_load1_pd(invsqrta+inr+0);
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm_setzero_pd();
 153         vgbsum           = _mm_setzero_pd();
 154         vvdwsum          = _mm_setzero_pd();
 155         dvdasum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 196                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 197
 198             /* Calculate table index by multiplying r with table scale and truncate to integer */
 199             rt               = _mm_mul_pd(r00,vftabscale);
 200             vfitab           = _mm_cvttpd_epi32(rt);
 201             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 202             vfitab           = _mm_slli_epi32(vfitab,3);
 203
 204             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 205             isaprod          = _mm_mul_pd(isai0,isaj0);
 206             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 207             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 208
 209             /* Calculate generalized born table index - this is a separate table from the normal one,
 210              * but we use the same procedure by multiplying r with scale and truncating to integer.
 211              */
 212             rt               = _mm_mul_pd(r00,gbscale);
 213             gbitab           = _mm_cvttpd_epi32(rt);
 214             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 215             gbitab           = _mm_slli_epi32(gbitab,2);
 216
 217             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 218             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 219             GMX_MM_TRANSPOSE2_PD(Y,F);
 220             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 221             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 222             GMX_MM_TRANSPOSE2_PD(G,H);
 223             Heps             = _mm_mul_pd(gbeps,H);
 224             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 225             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 226             vgb              = _mm_mul_pd(gbqqfactor,VV);
 227
 228             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 229             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 230             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 231             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 232             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 233             velec            = _mm_mul_pd(qq00,rinv00);
 234             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 235
 236             /* CUBIC SPLINE TABLE DISPERSION */
 237             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Heps             = _mm_mul_pd(vfeps,H);
 244             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 245             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 246             vvdw6            = _mm_mul_pd(c6_00,VV);
 247             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 248             fvdw6            = _mm_mul_pd(c6_00,FF);
 249
 250             /* CUBIC SPLINE TABLE REPULSION */
 251             vfitab           = _mm_add_epi32(vfitab,ifour);
 252             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             GMX_MM_TRANSPOSE2_PD(Y,F);
 255             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 256             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 257             GMX_MM_TRANSPOSE2_PD(G,H);
 258             Heps             = _mm_mul_pd(vfeps,H);
 259             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 260             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 261             vvdw12           = _mm_mul_pd(c12_00,VV);
 262             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 263             fvdw12           = _mm_mul_pd(c12_00,FF);
 264             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 265             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm_add_pd(velecsum,velec);
 269             vgbsum           = _mm_add_pd(vgbsum,vgb);
 270             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_pd(felec,fvdw);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_pd(fscal,dx00);
 276             ty               = _mm_mul_pd(fscal,dy00);
 277             tz               = _mm_mul_pd(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_pd(fix0,tx);
 281             fiy0             = _mm_add_pd(fiy0,ty);
 282             fiz0             = _mm_add_pd(fiz0,tz);
 283
 284             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 285
 286             /* Inner loop uses 92 flops */
 287         }
 288
 289         if(jidx<j_index_end)
 290         {
 291
 292             jnrA             = jjnr[jidx];
 293             j_coord_offsetA  = DIM*jnrA;
 294
 295             /* load j atom coordinates */
 296             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 297                                               &jx0,&jy0,&jz0);
 298
 299             /* Calculate displacement vector */
 300             dx00             = _mm_sub_pd(ix0,jx0);
 301             dy00             = _mm_sub_pd(iy0,jy0);
 302             dz00             = _mm_sub_pd(iz0,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 306
 307             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 308
 309             /* Load parameters for j particles */
 310             jq0              = _mm_load_sd(charge+jnrA+0);
 311             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 312             vdwjidx0A        = 2*vdwtype[jnrA+0];
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r00              = _mm_mul_pd(rsq00,rinv00);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq00             = _mm_mul_pd(iq0,jq0);
 322             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 323
 324             /* Calculate table index by multiplying r with table scale and truncate to integer */
 325             rt               = _mm_mul_pd(r00,vftabscale);
 326             vfitab           = _mm_cvttpd_epi32(rt);
 327             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 328             vfitab           = _mm_slli_epi32(vfitab,3);
 329
 330             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 331             isaprod          = _mm_mul_pd(isai0,isaj0);
 332             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 333             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 334
 335             /* Calculate generalized born table index - this is a separate table from the normal one,
 336              * but we use the same procedure by multiplying r with scale and truncating to integer.
 337              */
 338             rt               = _mm_mul_pd(r00,gbscale);
 339             gbitab           = _mm_cvttpd_epi32(rt);
 340             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 341             gbitab           = _mm_slli_epi32(gbitab,2);
 342
 343             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 344             F                = _mm_setzero_pd();
 345             GMX_MM_TRANSPOSE2_PD(Y,F);
 346             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 347             H                = _mm_setzero_pd();
 348             GMX_MM_TRANSPOSE2_PD(G,H);
 349             Heps             = _mm_mul_pd(gbeps,H);
 350             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 351             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 352             vgb              = _mm_mul_pd(gbqqfactor,VV);
 353
 354             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 355             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 356             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 357             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 358             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 359             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 360             velec            = _mm_mul_pd(qq00,rinv00);
 361             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 362
 363             /* CUBIC SPLINE TABLE DISPERSION */
 364             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 365             F                = _mm_setzero_pd();
 366             GMX_MM_TRANSPOSE2_PD(Y,F);
 367             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 368             H                = _mm_setzero_pd();
 369             GMX_MM_TRANSPOSE2_PD(G,H);
 370             Heps             = _mm_mul_pd(vfeps,H);
 371             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 372             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 373             vvdw6            = _mm_mul_pd(c6_00,VV);
 374             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 375             fvdw6            = _mm_mul_pd(c6_00,FF);
 376
 377             /* CUBIC SPLINE TABLE REPULSION */
 378             vfitab           = _mm_add_epi32(vfitab,ifour);
 379             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 380             F                = _mm_setzero_pd();
 381             GMX_MM_TRANSPOSE2_PD(Y,F);
 382             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 383             H                = _mm_setzero_pd();
 384             GMX_MM_TRANSPOSE2_PD(G,H);
 385             Heps             = _mm_mul_pd(vfeps,H);
 386             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 387             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 388             vvdw12           = _mm_mul_pd(c12_00,VV);
 389             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 390             fvdw12           = _mm_mul_pd(c12_00,FF);
 391             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 392             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 398             vgbsum           = _mm_add_pd(vgbsum,vgb);
 399             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 400             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 401
 402             fscal            = _mm_add_pd(felec,fvdw);
 403
 404             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm_mul_pd(fscal,dx00);
 408             ty               = _mm_mul_pd(fscal,dy00);
 409             tz               = _mm_mul_pd(fscal,dz00);
 410
 411             /* Update vectorial force */
 412             fix0             = _mm_add_pd(fix0,tx);
 413             fiy0             = _mm_add_pd(fiy0,ty);
 414             fiz0             = _mm_add_pd(fiz0,tz);
 415
 416             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 417
 418             /* Inner loop uses 92 flops */
 419         }
 420
 421         /* End of innermost loop */
 422
 423         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 424                                               f+i_coord_offset,fshift+i_shift_offset);
 425
 426         ggid                        = gid[iidx];
 427         /* Update potential energies */
 428         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 429         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 430         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 431         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 432         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 433
 434         /* Increment number of inner iterations */
 435         inneriter                  += j_index_end - j_index_start;
 436
 437         /* Outer loop uses 10 flops */
 438     }
 439
 440     /* Increment number of outer iterations */
 441     outeriter        += nri;
 442
 443     /* Update outer/inner flops */
 444
 445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
 446 }
 447 /*
 448  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
 449  * Electrostatics interaction: GeneralizedBorn
 450  * VdW interaction:            CubicSplineTable
 451  * Geometry:                   Particle-Particle
 452  * Calculate force/pot:        Force
 453  */
 454 void
 455 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
 456                     (t_nblist * gmx_restrict                nlist,
 457                      rvec * gmx_restrict                    xx,
 458                      rvec * gmx_restrict                    ff,
 459                      t_forcerec * gmx_restrict              fr,
 460                      t_mdatoms * gmx_restrict               mdatoms,
 461                      nb_kernel_data_t * gmx_restrict        kernel_data,
 462                      t_nrnb * gmx_restrict                  nrnb)
 463 {
 464     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 465      * just 0 for non-waters.
 466      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 467      * jnr indices corresponding to data put in the four positions in the SIMD register.
 468      */
 469     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 470     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 471     int              jnrA,jnrB;
 472     int              j_coord_offsetA,j_coord_offsetB;
 473     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 474     real             rcutoff_scalar;
 475     real             *shiftvec,*fshift,*x,*f;
 476     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 477     int              vdwioffset0;
 478     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 479     int              vdwjidx0A,vdwjidx0B;
 480     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 481     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 482     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 483     real             *charge;
 484     __m128i          gbitab;
 485     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 486     __m128d          minushalf = _mm_set1_pd(-0.5);
 487     real             *invsqrta,*dvda,*gbtab;
 488     int              nvdwtype;
 489     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 490     int              *vdwtype;
 491     real             *vdwparam;
 492     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 493     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 494     __m128i          vfitab;
 495     __m128i          ifour       = _mm_set1_epi32(4);
 496     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 497     real             *vftab;
 498     __m128d          dummy_mask,cutoff_mask;
 499     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 500     __m128d          one     = _mm_set1_pd(1.0);
 501     __m128d          two     = _mm_set1_pd(2.0);
 502     x                = xx[0];
 503     f                = ff[0];
 504
 505     nri              = nlist->nri;
 506     iinr             = nlist->iinr;
 507     jindex           = nlist->jindex;
 508     jjnr             = nlist->jjnr;
 509     shiftidx         = nlist->shift;
 510     gid              = nlist->gid;
 511     shiftvec         = fr->shift_vec[0];
 512     fshift           = fr->fshift[0];
 513     facel            = _mm_set1_pd(fr->epsfac);
 514     charge           = mdatoms->chargeA;
 515     nvdwtype         = fr->ntype;
 516     vdwparam         = fr->nbfp;
 517     vdwtype          = mdatoms->typeA;
 518
 519     vftab            = kernel_data->table_vdw->data;
 520     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 521
 522     invsqrta         = fr->invsqrta;
 523     dvda             = fr->dvda;
 524     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 525     gbtab            = fr->gbtab.data;
 526     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 527
 528     /* Avoid stupid compiler warnings */
 529     jnrA = jnrB = 0;
 530     j_coord_offsetA = 0;
 531     j_coord_offsetB = 0;
 532
 533     outeriter        = 0;
 534     inneriter        = 0;
 535
 536     /* Start outer loop over neighborlists */
 537     for(iidx=0; iidx<nri; iidx++)
 538     {
 539         /* Load shift vector for this list */
 540         i_shift_offset   = DIM*shiftidx[iidx];
 541
 542         /* Load limits for loop over neighbors */
 543         j_index_start    = jindex[iidx];
 544         j_index_end      = jindex[iidx+1];
 545
 546         /* Get outer coordinate index */
 547         inr              = iinr[iidx];
 548         i_coord_offset   = DIM*inr;
 549
 550         /* Load i particle coords and add shift vector */
 551         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 552
 553         fix0             = _mm_setzero_pd();
 554         fiy0             = _mm_setzero_pd();
 555         fiz0             = _mm_setzero_pd();
 556
 557         /* Load parameters for i particles */
 558         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 559         isai0            = _mm_load1_pd(invsqrta+inr+0);
 560         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 561
 562         dvdasum          = _mm_setzero_pd();
 563
 564         /* Start inner kernel loop */
 565         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 566         {
 567
 568             /* Get j neighbor index, and coordinate index */
 569             jnrA             = jjnr[jidx];
 570             jnrB             = jjnr[jidx+1];
 571             j_coord_offsetA  = DIM*jnrA;
 572             j_coord_offsetB  = DIM*jnrB;
 573
 574             /* load j atom coordinates */
 575             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 576                                               &jx0,&jy0,&jz0);
 577
 578             /* Calculate displacement vector */
 579             dx00             = _mm_sub_pd(ix0,jx0);
 580             dy00             = _mm_sub_pd(iy0,jy0);
 581             dz00             = _mm_sub_pd(iz0,jz0);
 582
 583             /* Calculate squared distance and things based on it */
 584             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 585
 586             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 587
 588             /* Load parameters for j particles */
 589             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 590             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 591             vdwjidx0A        = 2*vdwtype[jnrA+0];
 592             vdwjidx0B        = 2*vdwtype[jnrB+0];
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r00              = _mm_mul_pd(rsq00,rinv00);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq00             = _mm_mul_pd(iq0,jq0);
 602             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 603                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 604
 605             /* Calculate table index by multiplying r with table scale and truncate to integer */
 606             rt               = _mm_mul_pd(r00,vftabscale);
 607             vfitab           = _mm_cvttpd_epi32(rt);
 608             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 609             vfitab           = _mm_slli_epi32(vfitab,3);
 610
 611             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 612             isaprod          = _mm_mul_pd(isai0,isaj0);
 613             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 614             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 615
 616             /* Calculate generalized born table index - this is a separate table from the normal one,
 617              * but we use the same procedure by multiplying r with scale and truncating to integer.
 618              */
 619             rt               = _mm_mul_pd(r00,gbscale);
 620             gbitab           = _mm_cvttpd_epi32(rt);
 621             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 622             gbitab           = _mm_slli_epi32(gbitab,2);
 623
 624             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 625             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 626             GMX_MM_TRANSPOSE2_PD(Y,F);
 627             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 628             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 629             GMX_MM_TRANSPOSE2_PD(G,H);
 630             Heps             = _mm_mul_pd(gbeps,H);
 631             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 632             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 633             vgb              = _mm_mul_pd(gbqqfactor,VV);
 634
 635             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 636             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 637             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 638             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 639             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 640             velec            = _mm_mul_pd(qq00,rinv00);
 641             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 642
 643             /* CUBIC SPLINE TABLE DISPERSION */
 644             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 645             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 646             GMX_MM_TRANSPOSE2_PD(Y,F);
 647             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 648             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 649             GMX_MM_TRANSPOSE2_PD(G,H);
 650             Heps             = _mm_mul_pd(vfeps,H);
 651             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 652             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 653             fvdw6            = _mm_mul_pd(c6_00,FF);
 654
 655             /* CUBIC SPLINE TABLE REPULSION */
 656             vfitab           = _mm_add_epi32(vfitab,ifour);
 657             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 658             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 659             GMX_MM_TRANSPOSE2_PD(Y,F);
 660             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 661             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 662             GMX_MM_TRANSPOSE2_PD(G,H);
 663             Heps             = _mm_mul_pd(vfeps,H);
 664             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 665             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 666             fvdw12           = _mm_mul_pd(c12_00,FF);
 667             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 668
 669             fscal            = _mm_add_pd(felec,fvdw);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm_mul_pd(fscal,dx00);
 673             ty               = _mm_mul_pd(fscal,dy00);
 674             tz               = _mm_mul_pd(fscal,dz00);
 675
 676             /* Update vectorial force */
 677             fix0             = _mm_add_pd(fix0,tx);
 678             fiy0             = _mm_add_pd(fiy0,ty);
 679             fiz0             = _mm_add_pd(fiz0,tz);
 680
 681             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 682
 683             /* Inner loop uses 82 flops */
 684         }
 685
 686         if(jidx<j_index_end)
 687         {
 688
 689             jnrA             = jjnr[jidx];
 690             j_coord_offsetA  = DIM*jnrA;
 691
 692             /* load j atom coordinates */
 693             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 694                                               &jx0,&jy0,&jz0);
 695
 696             /* Calculate displacement vector */
 697             dx00             = _mm_sub_pd(ix0,jx0);
 698             dy00             = _mm_sub_pd(iy0,jy0);
 699             dz00             = _mm_sub_pd(iz0,jz0);
 700
 701             /* Calculate squared distance and things based on it */
 702             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 703
 704             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 705
 706             /* Load parameters for j particles */
 707             jq0              = _mm_load_sd(charge+jnrA+0);
 708             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 709             vdwjidx0A        = 2*vdwtype[jnrA+0];
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r00              = _mm_mul_pd(rsq00,rinv00);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq00             = _mm_mul_pd(iq0,jq0);
 719             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 720
 721             /* Calculate table index by multiplying r with table scale and truncate to integer */
 722             rt               = _mm_mul_pd(r00,vftabscale);
 723             vfitab           = _mm_cvttpd_epi32(rt);
 724             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 725             vfitab           = _mm_slli_epi32(vfitab,3);
 726
 727             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 728             isaprod          = _mm_mul_pd(isai0,isaj0);
 729             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 730             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 731
 732             /* Calculate generalized born table index - this is a separate table from the normal one,
 733              * but we use the same procedure by multiplying r with scale and truncating to integer.
 734              */
 735             rt               = _mm_mul_pd(r00,gbscale);
 736             gbitab           = _mm_cvttpd_epi32(rt);
 737             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 738             gbitab           = _mm_slli_epi32(gbitab,2);
 739
 740             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 741             F                = _mm_setzero_pd();
 742             GMX_MM_TRANSPOSE2_PD(Y,F);
 743             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 744             H                = _mm_setzero_pd();
 745             GMX_MM_TRANSPOSE2_PD(G,H);
 746             Heps             = _mm_mul_pd(gbeps,H);
 747             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 748             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 749             vgb              = _mm_mul_pd(gbqqfactor,VV);
 750
 751             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 752             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 753             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 754             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 755             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 756             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 757             velec            = _mm_mul_pd(qq00,rinv00);
 758             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 759
 760             /* CUBIC SPLINE TABLE DISPERSION */
 761             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 762             F                = _mm_setzero_pd();
 763             GMX_MM_TRANSPOSE2_PD(Y,F);
 764             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 765             H                = _mm_setzero_pd();
 766             GMX_MM_TRANSPOSE2_PD(G,H);
 767             Heps             = _mm_mul_pd(vfeps,H);
 768             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 769             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 770             fvdw6            = _mm_mul_pd(c6_00,FF);
 771
 772             /* CUBIC SPLINE TABLE REPULSION */
 773             vfitab           = _mm_add_epi32(vfitab,ifour);
 774             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 775             F                = _mm_setzero_pd();
 776             GMX_MM_TRANSPOSE2_PD(Y,F);
 777             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 778             H                = _mm_setzero_pd();
 779             GMX_MM_TRANSPOSE2_PD(G,H);
 780             Heps             = _mm_mul_pd(vfeps,H);
 781             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 782             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 783             fvdw12           = _mm_mul_pd(c12_00,FF);
 784             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 785
 786             fscal            = _mm_add_pd(felec,fvdw);
 787
 788             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 789
 790             /* Calculate temporary vectorial force */
 791             tx               = _mm_mul_pd(fscal,dx00);
 792             ty               = _mm_mul_pd(fscal,dy00);
 793             tz               = _mm_mul_pd(fscal,dz00);
 794
 795             /* Update vectorial force */
 796             fix0             = _mm_add_pd(fix0,tx);
 797             fiy0             = _mm_add_pd(fiy0,ty);
 798             fiz0             = _mm_add_pd(fiz0,tz);
 799
 800             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 801
 802             /* Inner loop uses 82 flops */
 803         }
 804
 805         /* End of innermost loop */
 806
 807         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 808                                               f+i_coord_offset,fshift+i_shift_offset);
 809
 810         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 811         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 812
 813         /* Increment number of inner iterations */
 814         inneriter                  += j_index_end - j_index_start;
 815
 816         /* Outer loop uses 7 flops */
 817     }
 818
 819     /* Increment number of outer iterations */
 820     outeriter        += nri;
 821
 822     /* Update outer/inner flops */
 823
 824     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);
 825 }