src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012, by the GROMACS development team, led by
   5  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   6  * others, as listed in the AUTHORS file in the top-level source
   7  * 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_double 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 "gmx_math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  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_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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          gbitab;
  90     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  91     __m128d          minushalf = _mm_set1_pd(-0.5);
  92     real             *invsqrta,*dvda,*gbtab;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     invsqrta         = fr->invsqrta;
 125     dvda             = fr->dvda;
 126     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 127     gbtab            = fr->gbtab.data;
 128     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         isai0            = _mm_load1_pd(invsqrta+inr+0);
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vgbsum           = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168         dvdasum          = _mm_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_pd(ix0,jx0);
 186             dy00             = _mm_sub_pd(iy0,jy0);
 187             dz00             = _mm_sub_pd(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 193
 194             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 199             vdwjidx0A        = 2*vdwtype[jnrA+0];
 200             vdwjidx0B        = 2*vdwtype[jnrB+0];
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             r00              = _mm_mul_pd(rsq00,rinv00);
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm_mul_pd(iq0,jq0);
 210             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 211                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 212
 213             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 214             isaprod          = _mm_mul_pd(isai0,isaj0);
 215             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 216             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 217
 218             /* Calculate generalized born table index - this is a separate table from the normal one,
 219              * but we use the same procedure by multiplying r with scale and truncating to integer.
 220              */
 221             rt               = _mm_mul_pd(r00,gbscale);
 222             gbitab           = _mm_cvttpd_epi32(rt);
 223             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 224             gbitab           = _mm_slli_epi32(gbitab,2);
 225
 226             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 227             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(Y,F);
 229             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 230             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(G,H);
 232             Heps             = _mm_mul_pd(gbeps,H);
 233             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 234             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 235             vgb              = _mm_mul_pd(gbqqfactor,VV);
 236
 237             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 238             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 239             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 240             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 241             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 242             velec            = _mm_mul_pd(qq00,rinv00);
 243             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 249             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 250             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 251             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 252
 253             /* Update potential sum for this i atom from the interaction with this j atom. */
 254             velecsum         = _mm_add_pd(velecsum,velec);
 255             vgbsum           = _mm_add_pd(vgbsum,vgb);
 256             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 257
 258             fscal            = _mm_add_pd(felec,fvdw);
 259
 260             /* Calculate temporary vectorial force */
 261             tx               = _mm_mul_pd(fscal,dx00);
 262             ty               = _mm_mul_pd(fscal,dy00);
 263             tz               = _mm_mul_pd(fscal,dz00);
 264
 265             /* Update vectorial force */
 266             fix0             = _mm_add_pd(fix0,tx);
 267             fiy0             = _mm_add_pd(fiy0,ty);
 268             fiz0             = _mm_add_pd(fiz0,tz);
 269
 270             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 271
 272             /* Inner loop uses 71 flops */
 273         }
 274
 275         if(jidx<j_index_end)
 276         {
 277
 278             jnrA             = jjnr[jidx];
 279             j_coord_offsetA  = DIM*jnrA;
 280
 281             /* load j atom coordinates */
 282             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 283                                               &jx0,&jy0,&jz0);
 284
 285             /* Calculate displacement vector */
 286             dx00             = _mm_sub_pd(ix0,jx0);
 287             dy00             = _mm_sub_pd(iy0,jy0);
 288             dz00             = _mm_sub_pd(iz0,jz0);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 292
 293             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 294
 295             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 296
 297             /* Load parameters for j particles */
 298             jq0              = _mm_load_sd(charge+jnrA+0);
 299             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r00              = _mm_mul_pd(rsq00,rinv00);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq00             = _mm_mul_pd(iq0,jq0);
 310             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 311
 312             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 313             isaprod          = _mm_mul_pd(isai0,isaj0);
 314             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 315             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 316
 317             /* Calculate generalized born table index - this is a separate table from the normal one,
 318              * but we use the same procedure by multiplying r with scale and truncating to integer.
 319              */
 320             rt               = _mm_mul_pd(r00,gbscale);
 321             gbitab           = _mm_cvttpd_epi32(rt);
 322             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 323             gbitab           = _mm_slli_epi32(gbitab,2);
 324
 325             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 326             F                = _mm_setzero_pd();
 327             GMX_MM_TRANSPOSE2_PD(Y,F);
 328             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 329             H                = _mm_setzero_pd();
 330             GMX_MM_TRANSPOSE2_PD(G,H);
 331             Heps             = _mm_mul_pd(gbeps,H);
 332             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 333             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 334             vgb              = _mm_mul_pd(gbqqfactor,VV);
 335
 336             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 337             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 338             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 339             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 340             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 341             velec            = _mm_mul_pd(qq00,rinv00);
 342             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 343
 344             /* LENNARD-JONES DISPERSION/REPULSION */
 345
 346             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 347             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 348             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 349             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 350             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 354             velecsum         = _mm_add_pd(velecsum,velec);
 355             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 356             vgbsum           = _mm_add_pd(vgbsum,vgb);
 357             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 358             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 359
 360             fscal            = _mm_add_pd(felec,fvdw);
 361
 362             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_pd(fscal,dx00);
 366             ty               = _mm_mul_pd(fscal,dy00);
 367             tz               = _mm_mul_pd(fscal,dz00);
 368
 369             /* Update vectorial force */
 370             fix0             = _mm_add_pd(fix0,tx);
 371             fiy0             = _mm_add_pd(fiy0,ty);
 372             fiz0             = _mm_add_pd(fiz0,tz);
 373
 374             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 375
 376             /* Inner loop uses 71 flops */
 377         }
 378
 379         /* End of innermost loop */
 380
 381         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 382                                               f+i_coord_offset,fshift+i_shift_offset);
 383
 384         ggid                        = gid[iidx];
 385         /* Update potential energies */
 386         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 387         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 388         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 389         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 390         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 391
 392         /* Increment number of inner iterations */
 393         inneriter                  += j_index_end - j_index_start;
 394
 395         /* Outer loop uses 10 flops */
 396     }
 397
 398     /* Increment number of outer iterations */
 399     outeriter        += nri;
 400
 401     /* Update outer/inner flops */
 402
 403     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 404 }
 405 /*
 406  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 407  * Electrostatics interaction: GeneralizedBorn
 408  * VdW interaction:            LennardJones
 409  * Geometry:                   Particle-Particle
 410  * Calculate force/pot:        Force
 411  */
 412 void
 413 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 414                     (t_nblist * gmx_restrict                nlist,
 415                      rvec * gmx_restrict                    xx,
 416                      rvec * gmx_restrict                    ff,
 417                      t_forcerec * gmx_restrict              fr,
 418                      t_mdatoms * gmx_restrict               mdatoms,
 419                      nb_kernel_data_t * gmx_restrict        kernel_data,
 420                      t_nrnb * gmx_restrict                  nrnb)
 421 {
 422     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 423      * just 0 for non-waters.
 424      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 425      * jnr indices corresponding to data put in the four positions in the SIMD register.
 426      */
 427     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 428     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 429     int              jnrA,jnrB;
 430     int              j_coord_offsetA,j_coord_offsetB;
 431     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 432     real             rcutoff_scalar;
 433     real             *shiftvec,*fshift,*x,*f;
 434     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 435     int              vdwioffset0;
 436     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 437     int              vdwjidx0A,vdwjidx0B;
 438     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 439     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 440     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 441     real             *charge;
 442     __m128i          gbitab;
 443     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 444     __m128d          minushalf = _mm_set1_pd(-0.5);
 445     real             *invsqrta,*dvda,*gbtab;
 446     int              nvdwtype;
 447     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 448     int              *vdwtype;
 449     real             *vdwparam;
 450     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 451     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 452     __m128i          vfitab;
 453     __m128i          ifour       = _mm_set1_epi32(4);
 454     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 455     real             *vftab;
 456     __m128d          dummy_mask,cutoff_mask;
 457     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 458     __m128d          one     = _mm_set1_pd(1.0);
 459     __m128d          two     = _mm_set1_pd(2.0);
 460     x                = xx[0];
 461     f                = ff[0];
 462
 463     nri              = nlist->nri;
 464     iinr             = nlist->iinr;
 465     jindex           = nlist->jindex;
 466     jjnr             = nlist->jjnr;
 467     shiftidx         = nlist->shift;
 468     gid              = nlist->gid;
 469     shiftvec         = fr->shift_vec[0];
 470     fshift           = fr->fshift[0];
 471     facel            = _mm_set1_pd(fr->epsfac);
 472     charge           = mdatoms->chargeA;
 473     nvdwtype         = fr->ntype;
 474     vdwparam         = fr->nbfp;
 475     vdwtype          = mdatoms->typeA;
 476
 477     invsqrta         = fr->invsqrta;
 478     dvda             = fr->dvda;
 479     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 480     gbtab            = fr->gbtab.data;
 481     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 482
 483     /* Avoid stupid compiler warnings */
 484     jnrA = jnrB = 0;
 485     j_coord_offsetA = 0;
 486     j_coord_offsetB = 0;
 487
 488     outeriter        = 0;
 489     inneriter        = 0;
 490
 491     /* Start outer loop over neighborlists */
 492     for(iidx=0; iidx<nri; iidx++)
 493     {
 494         /* Load shift vector for this list */
 495         i_shift_offset   = DIM*shiftidx[iidx];
 496
 497         /* Load limits for loop over neighbors */
 498         j_index_start    = jindex[iidx];
 499         j_index_end      = jindex[iidx+1];
 500
 501         /* Get outer coordinate index */
 502         inr              = iinr[iidx];
 503         i_coord_offset   = DIM*inr;
 504
 505         /* Load i particle coords and add shift vector */
 506         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 507
 508         fix0             = _mm_setzero_pd();
 509         fiy0             = _mm_setzero_pd();
 510         fiz0             = _mm_setzero_pd();
 511
 512         /* Load parameters for i particles */
 513         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 514         isai0            = _mm_load1_pd(invsqrta+inr+0);
 515         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 516
 517         dvdasum          = _mm_setzero_pd();
 518
 519         /* Start inner kernel loop */
 520         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 521         {
 522
 523             /* Get j neighbor index, and coordinate index */
 524             jnrA             = jjnr[jidx];
 525             jnrB             = jjnr[jidx+1];
 526             j_coord_offsetA  = DIM*jnrA;
 527             j_coord_offsetB  = DIM*jnrB;
 528
 529             /* load j atom coordinates */
 530             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                               &jx0,&jy0,&jz0);
 532
 533             /* Calculate displacement vector */
 534             dx00             = _mm_sub_pd(ix0,jx0);
 535             dy00             = _mm_sub_pd(iy0,jy0);
 536             dz00             = _mm_sub_pd(iz0,jz0);
 537
 538             /* Calculate squared distance and things based on it */
 539             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 540
 541             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 542
 543             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 544
 545             /* Load parameters for j particles */
 546             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 547             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 548             vdwjidx0A        = 2*vdwtype[jnrA+0];
 549             vdwjidx0B        = 2*vdwtype[jnrB+0];
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r00              = _mm_mul_pd(rsq00,rinv00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_pd(iq0,jq0);
 559             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 561
 562             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 563             isaprod          = _mm_mul_pd(isai0,isaj0);
 564             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 565             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 566
 567             /* Calculate generalized born table index - this is a separate table from the normal one,
 568              * but we use the same procedure by multiplying r with scale and truncating to integer.
 569              */
 570             rt               = _mm_mul_pd(r00,gbscale);
 571             gbitab           = _mm_cvttpd_epi32(rt);
 572             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 573             gbitab           = _mm_slli_epi32(gbitab,2);
 574
 575             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 576             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 577             GMX_MM_TRANSPOSE2_PD(Y,F);
 578             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 579             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 580             GMX_MM_TRANSPOSE2_PD(G,H);
 581             Heps             = _mm_mul_pd(gbeps,H);
 582             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 583             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 584             vgb              = _mm_mul_pd(gbqqfactor,VV);
 585
 586             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 587             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 588             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 589             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 590             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 591             velec            = _mm_mul_pd(qq00,rinv00);
 592             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 593
 594             /* LENNARD-JONES DISPERSION/REPULSION */
 595
 596             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 597             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 598
 599             fscal            = _mm_add_pd(felec,fvdw);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_pd(fscal,dx00);
 603             ty               = _mm_mul_pd(fscal,dy00);
 604             tz               = _mm_mul_pd(fscal,dz00);
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_add_pd(fix0,tx);
 608             fiy0             = _mm_add_pd(fiy0,ty);
 609             fiz0             = _mm_add_pd(fiz0,tz);
 610
 611             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 612
 613             /* Inner loop uses 64 flops */
 614         }
 615
 616         if(jidx<j_index_end)
 617         {
 618
 619             jnrA             = jjnr[jidx];
 620             j_coord_offsetA  = DIM*jnrA;
 621
 622             /* load j atom coordinates */
 623             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 624                                               &jx0,&jy0,&jz0);
 625
 626             /* Calculate displacement vector */
 627             dx00             = _mm_sub_pd(ix0,jx0);
 628             dy00             = _mm_sub_pd(iy0,jy0);
 629             dz00             = _mm_sub_pd(iz0,jz0);
 630
 631             /* Calculate squared distance and things based on it */
 632             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 633
 634             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 635
 636             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 637
 638             /* Load parameters for j particles */
 639             jq0              = _mm_load_sd(charge+jnrA+0);
 640             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 641             vdwjidx0A        = 2*vdwtype[jnrA+0];
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r00              = _mm_mul_pd(rsq00,rinv00);
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq00             = _mm_mul_pd(iq0,jq0);
 651             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 652
 653             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 654             isaprod          = _mm_mul_pd(isai0,isaj0);
 655             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 656             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 657
 658             /* Calculate generalized born table index - this is a separate table from the normal one,
 659              * but we use the same procedure by multiplying r with scale and truncating to integer.
 660              */
 661             rt               = _mm_mul_pd(r00,gbscale);
 662             gbitab           = _mm_cvttpd_epi32(rt);
 663             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 664             gbitab           = _mm_slli_epi32(gbitab,2);
 665
 666             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 667             F                = _mm_setzero_pd();
 668             GMX_MM_TRANSPOSE2_PD(Y,F);
 669             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 670             H                = _mm_setzero_pd();
 671             GMX_MM_TRANSPOSE2_PD(G,H);
 672             Heps             = _mm_mul_pd(gbeps,H);
 673             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 674             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 675             vgb              = _mm_mul_pd(gbqqfactor,VV);
 676
 677             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 678             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 679             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 680             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 681             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 682             velec            = _mm_mul_pd(qq00,rinv00);
 683             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 684
 685             /* LENNARD-JONES DISPERSION/REPULSION */
 686
 687             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 688             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 689
 690             fscal            = _mm_add_pd(felec,fvdw);
 691
 692             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm_mul_pd(fscal,dx00);
 696             ty               = _mm_mul_pd(fscal,dy00);
 697             tz               = _mm_mul_pd(fscal,dz00);
 698
 699             /* Update vectorial force */
 700             fix0             = _mm_add_pd(fix0,tx);
 701             fiy0             = _mm_add_pd(fiy0,ty);
 702             fiz0             = _mm_add_pd(fiz0,tz);
 703
 704             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 705
 706             /* Inner loop uses 64 flops */
 707         }
 708
 709         /* End of innermost loop */
 710
 711         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 712                                               f+i_coord_offset,fshift+i_shift_offset);
 713
 714         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 715         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 716
 717         /* Increment number of inner iterations */
 718         inneriter                  += j_index_end - j_index_start;
 719
 720         /* Outer loop uses 7 flops */
 721     }
 722
 723     /* Increment number of outer iterations */
 724     outeriter        += nri;
 725
 726     /* Update outer/inner flops */
 727
 728     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 729 }