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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_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_sse2_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_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B 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_cvtepi32_pd(gbitab));
 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_cvtepi32_pd(gbitab));
 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             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 340             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 341             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 342             velec            = _mm_mul_pd(qq00,rinv00);
 343             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 344
 345             /* LENNARD-JONES DISPERSION/REPULSION */
 346
 347             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 348             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 349             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 350             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 351             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 357             vgbsum           = _mm_add_pd(vgbsum,vgb);
 358             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 359             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 360
 361             fscal            = _mm_add_pd(felec,fvdw);
 362
 363             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_pd(fscal,dx00);
 367             ty               = _mm_mul_pd(fscal,dy00);
 368             tz               = _mm_mul_pd(fscal,dz00);
 369
 370             /* Update vectorial force */
 371             fix0             = _mm_add_pd(fix0,tx);
 372             fiy0             = _mm_add_pd(fiy0,ty);
 373             fiz0             = _mm_add_pd(fiz0,tz);
 374
 375             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 376
 377             /* Inner loop uses 71 flops */
 378         }
 379
 380         /* End of innermost loop */
 381
 382         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 383                                               f+i_coord_offset,fshift+i_shift_offset);
 384
 385         ggid                        = gid[iidx];
 386         /* Update potential energies */
 387         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 388         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 389         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 390         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 391         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 392
 393         /* Increment number of inner iterations */
 394         inneriter                  += j_index_end - j_index_start;
 395
 396         /* Outer loop uses 10 flops */
 397     }
 398
 399     /* Increment number of outer iterations */
 400     outeriter        += nri;
 401
 402     /* Update outer/inner flops */
 403
 404     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 405 }
 406 /*
 407  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_double
 408  * Electrostatics interaction: GeneralizedBorn
 409  * VdW interaction:            LennardJones
 410  * Geometry:                   Particle-Particle
 411  * Calculate force/pot:        Force
 412  */
 413 void
 414 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_double
 415                     (t_nblist                    * gmx_restrict       nlist,
 416                      rvec                        * gmx_restrict          xx,
 417                      rvec                        * gmx_restrict          ff,
 418                      t_forcerec                  * gmx_restrict          fr,
 419                      t_mdatoms                   * gmx_restrict     mdatoms,
 420                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 421                      t_nrnb                      * gmx_restrict        nrnb)
 422 {
 423     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 424      * just 0 for non-waters.
 425      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 426      * jnr indices corresponding to data put in the four positions in the SIMD register.
 427      */
 428     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 429     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 430     int              jnrA,jnrB;
 431     int              j_coord_offsetA,j_coord_offsetB;
 432     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 433     real             rcutoff_scalar;
 434     real             *shiftvec,*fshift,*x,*f;
 435     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 436     int              vdwioffset0;
 437     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 438     int              vdwjidx0A,vdwjidx0B;
 439     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 440     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 441     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 442     real             *charge;
 443     __m128i          gbitab;
 444     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 445     __m128d          minushalf = _mm_set1_pd(-0.5);
 446     real             *invsqrta,*dvda,*gbtab;
 447     int              nvdwtype;
 448     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 449     int              *vdwtype;
 450     real             *vdwparam;
 451     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 452     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 453     __m128i          vfitab;
 454     __m128i          ifour       = _mm_set1_epi32(4);
 455     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 456     real             *vftab;
 457     __m128d          dummy_mask,cutoff_mask;
 458     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 459     __m128d          one     = _mm_set1_pd(1.0);
 460     __m128d          two     = _mm_set1_pd(2.0);
 461     x                = xx[0];
 462     f                = ff[0];
 463
 464     nri              = nlist->nri;
 465     iinr             = nlist->iinr;
 466     jindex           = nlist->jindex;
 467     jjnr             = nlist->jjnr;
 468     shiftidx         = nlist->shift;
 469     gid              = nlist->gid;
 470     shiftvec         = fr->shift_vec[0];
 471     fshift           = fr->fshift[0];
 472     facel            = _mm_set1_pd(fr->epsfac);
 473     charge           = mdatoms->chargeA;
 474     nvdwtype         = fr->ntype;
 475     vdwparam         = fr->nbfp;
 476     vdwtype          = mdatoms->typeA;
 477
 478     invsqrta         = fr->invsqrta;
 479     dvda             = fr->dvda;
 480     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 481     gbtab            = fr->gbtab.data;
 482     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 483
 484     /* Avoid stupid compiler warnings */
 485     jnrA = jnrB = 0;
 486     j_coord_offsetA = 0;
 487     j_coord_offsetB = 0;
 488
 489     outeriter        = 0;
 490     inneriter        = 0;
 491
 492     /* Start outer loop over neighborlists */
 493     for(iidx=0; iidx<nri; iidx++)
 494     {
 495         /* Load shift vector for this list */
 496         i_shift_offset   = DIM*shiftidx[iidx];
 497
 498         /* Load limits for loop over neighbors */
 499         j_index_start    = jindex[iidx];
 500         j_index_end      = jindex[iidx+1];
 501
 502         /* Get outer coordinate index */
 503         inr              = iinr[iidx];
 504         i_coord_offset   = DIM*inr;
 505
 506         /* Load i particle coords and add shift vector */
 507         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 508
 509         fix0             = _mm_setzero_pd();
 510         fiy0             = _mm_setzero_pd();
 511         fiz0             = _mm_setzero_pd();
 512
 513         /* Load parameters for i particles */
 514         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 515         isai0            = _mm_load1_pd(invsqrta+inr+0);
 516         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 517
 518         dvdasum          = _mm_setzero_pd();
 519
 520         /* Start inner kernel loop */
 521         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrA             = jjnr[jidx];
 526             jnrB             = jjnr[jidx+1];
 527             j_coord_offsetA  = DIM*jnrA;
 528             j_coord_offsetB  = DIM*jnrB;
 529
 530             /* load j atom coordinates */
 531             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm_sub_pd(ix0,jx0);
 536             dy00             = _mm_sub_pd(iy0,jy0);
 537             dz00             = _mm_sub_pd(iz0,jz0);
 538
 539             /* Calculate squared distance and things based on it */
 540             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 541
 542             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 543
 544             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 545
 546             /* Load parameters for j particles */
 547             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 548             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 549             vdwjidx0A        = 2*vdwtype[jnrA+0];
 550             vdwjidx0B        = 2*vdwtype[jnrB+0];
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             r00              = _mm_mul_pd(rsq00,rinv00);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq00             = _mm_mul_pd(iq0,jq0);
 560             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 562
 563             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 564             isaprod          = _mm_mul_pd(isai0,isaj0);
 565             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 566             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 567
 568             /* Calculate generalized born table index - this is a separate table from the normal one,
 569              * but we use the same procedure by multiplying r with scale and truncating to integer.
 570              */
 571             rt               = _mm_mul_pd(r00,gbscale);
 572             gbitab           = _mm_cvttpd_epi32(rt);
 573             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 574             gbitab           = _mm_slli_epi32(gbitab,2);
 575
 576             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 577             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 578             GMX_MM_TRANSPOSE2_PD(Y,F);
 579             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 580             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 581             GMX_MM_TRANSPOSE2_PD(G,H);
 582             Heps             = _mm_mul_pd(gbeps,H);
 583             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 584             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 585             vgb              = _mm_mul_pd(gbqqfactor,VV);
 586
 587             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 588             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 589             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 590             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 591             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 592             velec            = _mm_mul_pd(qq00,rinv00);
 593             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 594
 595             /* LENNARD-JONES DISPERSION/REPULSION */
 596
 597             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 598             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 599
 600             fscal            = _mm_add_pd(felec,fvdw);
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm_mul_pd(fscal,dx00);
 604             ty               = _mm_mul_pd(fscal,dy00);
 605             tz               = _mm_mul_pd(fscal,dz00);
 606
 607             /* Update vectorial force */
 608             fix0             = _mm_add_pd(fix0,tx);
 609             fiy0             = _mm_add_pd(fiy0,ty);
 610             fiz0             = _mm_add_pd(fiz0,tz);
 611
 612             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 613
 614             /* Inner loop uses 64 flops */
 615         }
 616
 617         if(jidx<j_index_end)
 618         {
 619
 620             jnrA             = jjnr[jidx];
 621             j_coord_offsetA  = DIM*jnrA;
 622
 623             /* load j atom coordinates */
 624             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 625                                               &jx0,&jy0,&jz0);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm_sub_pd(ix0,jx0);
 629             dy00             = _mm_sub_pd(iy0,jy0);
 630             dz00             = _mm_sub_pd(iz0,jz0);
 631
 632             /* Calculate squared distance and things based on it */
 633             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 634
 635             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 636
 637             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 638
 639             /* Load parameters for j particles */
 640             jq0              = _mm_load_sd(charge+jnrA+0);
 641             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 642             vdwjidx0A        = 2*vdwtype[jnrA+0];
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r00              = _mm_mul_pd(rsq00,rinv00);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq00             = _mm_mul_pd(iq0,jq0);
 652             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 653
 654             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 655             isaprod          = _mm_mul_pd(isai0,isaj0);
 656             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 657             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 658
 659             /* Calculate generalized born table index - this is a separate table from the normal one,
 660              * but we use the same procedure by multiplying r with scale and truncating to integer.
 661              */
 662             rt               = _mm_mul_pd(r00,gbscale);
 663             gbitab           = _mm_cvttpd_epi32(rt);
 664             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 665             gbitab           = _mm_slli_epi32(gbitab,2);
 666
 667             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 668             F                = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(Y,F);
 670             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 671             H                = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(G,H);
 673             Heps             = _mm_mul_pd(gbeps,H);
 674             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 675             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 676             vgb              = _mm_mul_pd(gbqqfactor,VV);
 677
 678             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 679             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 680             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 681             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 682             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 683             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 684             velec            = _mm_mul_pd(qq00,rinv00);
 685             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 686
 687             /* LENNARD-JONES DISPERSION/REPULSION */
 688
 689             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 690             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 691
 692             fscal            = _mm_add_pd(felec,fvdw);
 693
 694             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm_mul_pd(fscal,dx00);
 698             ty               = _mm_mul_pd(fscal,dy00);
 699             tz               = _mm_mul_pd(fscal,dz00);
 700
 701             /* Update vectorial force */
 702             fix0             = _mm_add_pd(fix0,tx);
 703             fiy0             = _mm_add_pd(fiy0,ty);
 704             fiz0             = _mm_add_pd(fiz0,tz);
 705
 706             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 707
 708             /* Inner loop uses 64 flops */
 709         }
 710
 711         /* End of innermost loop */
 712
 713         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 714                                               f+i_coord_offset,fshift+i_shift_offset);
 715
 716         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 717         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 718
 719         /* Increment number of inner iterations */
 720         inneriter                  += j_index_end - j_index_start;
 721
 722         /* Outer loop uses 7 flops */
 723     }
 724
 725     /* Increment number of outer iterations */
 726     outeriter        += nri;
 727
 728     /* Update outer/inner flops */
 729
 730     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 731 }