src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128i          gbitab;
  87     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  88     __m128d          minushalf = _mm_set1_pd(-0.5);
  89     real             *invsqrta,*dvda,*gbtab;
  90     int              nvdwtype;
  91     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  95     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     invsqrta         = fr->invsqrta;
 122     dvda             = fr->dvda;
 123     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 124     gbtab            = fr->gbtab->data;
 125     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 158         isai0            = _mm_load1_pd(invsqrta+inr+0);
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_pd();
 163         vgbsum           = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165         dvdasum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = sse2_invsqrt_d(rsq00);
 190
 191             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _mm_mul_pd(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_pd(iq0,jq0);
 207             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209
 210             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 211             isaprod          = _mm_mul_pd(isai0,isaj0);
 212             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 213             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 214
 215             /* Calculate generalized born table index - this is a separate table from the normal one,
 216              * but we use the same procedure by multiplying r with scale and truncating to integer.
 217              */
 218             rt               = _mm_mul_pd(r00,gbscale);
 219             gbitab           = _mm_cvttpd_epi32(rt);
 220             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 221             gbitab           = _mm_slli_epi32(gbitab,2);
 222
 223             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 224             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 225             GMX_MM_TRANSPOSE2_PD(Y,F);
 226             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 227             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 228             GMX_MM_TRANSPOSE2_PD(G,H);
 229             Heps             = _mm_mul_pd(gbeps,H);
 230             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 231             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 232             vgb              = _mm_mul_pd(gbqqfactor,VV);
 233
 234             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 235             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 236             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 237             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 238             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 239             velec            = _mm_mul_pd(qq00,rinv00);
 240             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 241
 242             /* LENNARD-JONES DISPERSION/REPULSION */
 243
 244             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 245             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 246             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 247             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 248             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm_add_pd(velecsum,velec);
 252             vgbsum           = _mm_add_pd(vgbsum,vgb);
 253             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 254
 255             fscal            = _mm_add_pd(felec,fvdw);
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm_mul_pd(fscal,dx00);
 259             ty               = _mm_mul_pd(fscal,dy00);
 260             tz               = _mm_mul_pd(fscal,dz00);
 261
 262             /* Update vectorial force */
 263             fix0             = _mm_add_pd(fix0,tx);
 264             fiy0             = _mm_add_pd(fiy0,ty);
 265             fiz0             = _mm_add_pd(fiz0,tz);
 266
 267             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 268
 269             /* Inner loop uses 71 flops */
 270         }
 271
 272         if(jidx<j_index_end)
 273         {
 274
 275             jnrA             = jjnr[jidx];
 276             j_coord_offsetA  = DIM*jnrA;
 277
 278             /* load j atom coordinates */
 279             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 280                                               &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm_sub_pd(ix0,jx0);
 284             dy00             = _mm_sub_pd(iy0,jy0);
 285             dz00             = _mm_sub_pd(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 289
 290             rinv00           = sse2_invsqrt_d(rsq00);
 291
 292             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 293
 294             /* Load parameters for j particles */
 295             jq0              = _mm_load_sd(charge+jnrA+0);
 296             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 297             vdwjidx0A        = 2*vdwtype[jnrA+0];
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r00              = _mm_mul_pd(rsq00,rinv00);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq00             = _mm_mul_pd(iq0,jq0);
 307             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 308
 309             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 310             isaprod          = _mm_mul_pd(isai0,isaj0);
 311             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 312             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 313
 314             /* Calculate generalized born table index - this is a separate table from the normal one,
 315              * but we use the same procedure by multiplying r with scale and truncating to integer.
 316              */
 317             rt               = _mm_mul_pd(r00,gbscale);
 318             gbitab           = _mm_cvttpd_epi32(rt);
 319             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 320             gbitab           = _mm_slli_epi32(gbitab,2);
 321
 322             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 323             F                = _mm_setzero_pd();
 324             GMX_MM_TRANSPOSE2_PD(Y,F);
 325             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 326             H                = _mm_setzero_pd();
 327             GMX_MM_TRANSPOSE2_PD(G,H);
 328             Heps             = _mm_mul_pd(gbeps,H);
 329             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 330             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 331             vgb              = _mm_mul_pd(gbqqfactor,VV);
 332
 333             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 334             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 335             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 336             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 337             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 338             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 339             velec            = _mm_mul_pd(qq00,rinv00);
 340             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 341
 342             /* LENNARD-JONES DISPERSION/REPULSION */
 343
 344             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 345             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 346             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 347             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 348             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 352             velecsum         = _mm_add_pd(velecsum,velec);
 353             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 354             vgbsum           = _mm_add_pd(vgbsum,vgb);
 355             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 356             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 357
 358             fscal            = _mm_add_pd(felec,fvdw);
 359
 360             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_pd(fscal,dx00);
 364             ty               = _mm_mul_pd(fscal,dy00);
 365             tz               = _mm_mul_pd(fscal,dz00);
 366
 367             /* Update vectorial force */
 368             fix0             = _mm_add_pd(fix0,tx);
 369             fiy0             = _mm_add_pd(fiy0,ty);
 370             fiz0             = _mm_add_pd(fiz0,tz);
 371
 372             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 373
 374             /* Inner loop uses 71 flops */
 375         }
 376
 377         /* End of innermost loop */
 378
 379         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 380                                               f+i_coord_offset,fshift+i_shift_offset);
 381
 382         ggid                        = gid[iidx];
 383         /* Update potential energies */
 384         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 385         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 386         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 387         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 388         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 389
 390         /* Increment number of inner iterations */
 391         inneriter                  += j_index_end - j_index_start;
 392
 393         /* Outer loop uses 10 flops */
 394     }
 395
 396     /* Increment number of outer iterations */
 397     outeriter        += nri;
 398
 399     /* Update outer/inner flops */
 400
 401     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 402 }
 403 /*
 404  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_double
 405  * Electrostatics interaction: GeneralizedBorn
 406  * VdW interaction:            LennardJones
 407  * Geometry:                   Particle-Particle
 408  * Calculate force/pot:        Force
 409  */
 410 void
 411 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_double
 412                     (t_nblist                    * gmx_restrict       nlist,
 413                      rvec                        * gmx_restrict          xx,
 414                      rvec                        * gmx_restrict          ff,
 415                      struct t_forcerec           * gmx_restrict          fr,
 416                      t_mdatoms                   * gmx_restrict     mdatoms,
 417                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 418                      t_nrnb                      * gmx_restrict        nrnb)
 419 {
 420     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 421      * just 0 for non-waters.
 422      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 423      * jnr indices corresponding to data put in the four positions in the SIMD register.
 424      */
 425     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 426     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 427     int              jnrA,jnrB;
 428     int              j_coord_offsetA,j_coord_offsetB;
 429     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 430     real             rcutoff_scalar;
 431     real             *shiftvec,*fshift,*x,*f;
 432     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 433     int              vdwioffset0;
 434     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 435     int              vdwjidx0A,vdwjidx0B;
 436     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 437     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 438     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 439     real             *charge;
 440     __m128i          gbitab;
 441     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 442     __m128d          minushalf = _mm_set1_pd(-0.5);
 443     real             *invsqrta,*dvda,*gbtab;
 444     int              nvdwtype;
 445     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 446     int              *vdwtype;
 447     real             *vdwparam;
 448     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 449     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 450     __m128i          vfitab;
 451     __m128i          ifour       = _mm_set1_epi32(4);
 452     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 453     real             *vftab;
 454     __m128d          dummy_mask,cutoff_mask;
 455     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 456     __m128d          one     = _mm_set1_pd(1.0);
 457     __m128d          two     = _mm_set1_pd(2.0);
 458     x                = xx[0];
 459     f                = ff[0];
 460
 461     nri              = nlist->nri;
 462     iinr             = nlist->iinr;
 463     jindex           = nlist->jindex;
 464     jjnr             = nlist->jjnr;
 465     shiftidx         = nlist->shift;
 466     gid              = nlist->gid;
 467     shiftvec         = fr->shift_vec[0];
 468     fshift           = fr->fshift[0];
 469     facel            = _mm_set1_pd(fr->ic->epsfac);
 470     charge           = mdatoms->chargeA;
 471     nvdwtype         = fr->ntype;
 472     vdwparam         = fr->nbfp;
 473     vdwtype          = mdatoms->typeA;
 474
 475     invsqrta         = fr->invsqrta;
 476     dvda             = fr->dvda;
 477     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 478     gbtab            = fr->gbtab->data;
 479     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 480
 481     /* Avoid stupid compiler warnings */
 482     jnrA = jnrB = 0;
 483     j_coord_offsetA = 0;
 484     j_coord_offsetB = 0;
 485
 486     outeriter        = 0;
 487     inneriter        = 0;
 488
 489     /* Start outer loop over neighborlists */
 490     for(iidx=0; iidx<nri; iidx++)
 491     {
 492         /* Load shift vector for this list */
 493         i_shift_offset   = DIM*shiftidx[iidx];
 494
 495         /* Load limits for loop over neighbors */
 496         j_index_start    = jindex[iidx];
 497         j_index_end      = jindex[iidx+1];
 498
 499         /* Get outer coordinate index */
 500         inr              = iinr[iidx];
 501         i_coord_offset   = DIM*inr;
 502
 503         /* Load i particle coords and add shift vector */
 504         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 505
 506         fix0             = _mm_setzero_pd();
 507         fiy0             = _mm_setzero_pd();
 508         fiz0             = _mm_setzero_pd();
 509
 510         /* Load parameters for i particles */
 511         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 512         isai0            = _mm_load1_pd(invsqrta+inr+0);
 513         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 514
 515         dvdasum          = _mm_setzero_pd();
 516
 517         /* Start inner kernel loop */
 518         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 519         {
 520
 521             /* Get j neighbor index, and coordinate index */
 522             jnrA             = jjnr[jidx];
 523             jnrB             = jjnr[jidx+1];
 524             j_coord_offsetA  = DIM*jnrA;
 525             j_coord_offsetB  = DIM*jnrB;
 526
 527             /* load j atom coordinates */
 528             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 529                                               &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm_sub_pd(ix0,jx0);
 533             dy00             = _mm_sub_pd(iy0,jy0);
 534             dz00             = _mm_sub_pd(iz0,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 538
 539             rinv00           = sse2_invsqrt_d(rsq00);
 540
 541             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 542
 543             /* Load parameters for j particles */
 544             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 545             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 546             vdwjidx0A        = 2*vdwtype[jnrA+0];
 547             vdwjidx0B        = 2*vdwtype[jnrB+0];
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             r00              = _mm_mul_pd(rsq00,rinv00);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq00             = _mm_mul_pd(iq0,jq0);
 557             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 558                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 559
 560             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 561             isaprod          = _mm_mul_pd(isai0,isaj0);
 562             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 563             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 564
 565             /* Calculate generalized born table index - this is a separate table from the normal one,
 566              * but we use the same procedure by multiplying r with scale and truncating to integer.
 567              */
 568             rt               = _mm_mul_pd(r00,gbscale);
 569             gbitab           = _mm_cvttpd_epi32(rt);
 570             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 571             gbitab           = _mm_slli_epi32(gbitab,2);
 572
 573             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 574             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 575             GMX_MM_TRANSPOSE2_PD(Y,F);
 576             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 577             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 578             GMX_MM_TRANSPOSE2_PD(G,H);
 579             Heps             = _mm_mul_pd(gbeps,H);
 580             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 581             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 582             vgb              = _mm_mul_pd(gbqqfactor,VV);
 583
 584             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 585             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 586             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 587             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 588             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 589             velec            = _mm_mul_pd(qq00,rinv00);
 590             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 591
 592             /* LENNARD-JONES DISPERSION/REPULSION */
 593
 594             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 595             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 596
 597             fscal            = _mm_add_pd(felec,fvdw);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm_mul_pd(fscal,dx00);
 601             ty               = _mm_mul_pd(fscal,dy00);
 602             tz               = _mm_mul_pd(fscal,dz00);
 603
 604             /* Update vectorial force */
 605             fix0             = _mm_add_pd(fix0,tx);
 606             fiy0             = _mm_add_pd(fiy0,ty);
 607             fiz0             = _mm_add_pd(fiz0,tz);
 608
 609             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 610
 611             /* Inner loop uses 64 flops */
 612         }
 613
 614         if(jidx<j_index_end)
 615         {
 616
 617             jnrA             = jjnr[jidx];
 618             j_coord_offsetA  = DIM*jnrA;
 619
 620             /* load j atom coordinates */
 621             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 622                                               &jx0,&jy0,&jz0);
 623
 624             /* Calculate displacement vector */
 625             dx00             = _mm_sub_pd(ix0,jx0);
 626             dy00             = _mm_sub_pd(iy0,jy0);
 627             dz00             = _mm_sub_pd(iz0,jz0);
 628
 629             /* Calculate squared distance and things based on it */
 630             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 631
 632             rinv00           = sse2_invsqrt_d(rsq00);
 633
 634             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 635
 636             /* Load parameters for j particles */
 637             jq0              = _mm_load_sd(charge+jnrA+0);
 638             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 639             vdwjidx0A        = 2*vdwtype[jnrA+0];
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             r00              = _mm_mul_pd(rsq00,rinv00);
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq00             = _mm_mul_pd(iq0,jq0);
 649             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 650
 651             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 652             isaprod          = _mm_mul_pd(isai0,isaj0);
 653             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 654             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 655
 656             /* Calculate generalized born table index - this is a separate table from the normal one,
 657              * but we use the same procedure by multiplying r with scale and truncating to integer.
 658              */
 659             rt               = _mm_mul_pd(r00,gbscale);
 660             gbitab           = _mm_cvttpd_epi32(rt);
 661             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 662             gbitab           = _mm_slli_epi32(gbitab,2);
 663
 664             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 665             F                = _mm_setzero_pd();
 666             GMX_MM_TRANSPOSE2_PD(Y,F);
 667             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 668             H                = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(G,H);
 670             Heps             = _mm_mul_pd(gbeps,H);
 671             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 672             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 673             vgb              = _mm_mul_pd(gbqqfactor,VV);
 674
 675             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 676             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 677             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 678             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 679             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 680             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 681             velec            = _mm_mul_pd(qq00,rinv00);
 682             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 683
 684             /* LENNARD-JONES DISPERSION/REPULSION */
 685
 686             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 687             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 688
 689             fscal            = _mm_add_pd(felec,fvdw);
 690
 691             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_pd(fscal,dx00);
 695             ty               = _mm_mul_pd(fscal,dy00);
 696             tz               = _mm_mul_pd(fscal,dz00);
 697
 698             /* Update vectorial force */
 699             fix0             = _mm_add_pd(fix0,tx);
 700             fiy0             = _mm_add_pd(fiy0,ty);
 701             fiz0             = _mm_add_pd(fiz0,tz);
 702
 703             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 704
 705             /* Inner loop uses 64 flops */
 706         }
 707
 708         /* End of innermost loop */
 709
 710         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 711                                               f+i_coord_offset,fshift+i_shift_offset);
 712
 713         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 714         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 7 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 728 }