src/gromacs/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) 2012,2013,2014, 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 sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: GeneralizedBorn
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128i          gbitab;
  88     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  89     __m128d          minushalf = _mm_set1_pd(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     int              nvdwtype;
  92     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  96     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  97     __m128i          vfitab;
  98     __m128i          ifour       = _mm_set1_epi32(4);
  99     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 100     real             *vftab;
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     invsqrta         = fr->invsqrta;
 123     dvda             = fr->dvda;
 124     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 125     gbtab            = fr->gbtab.data;
 126     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 159         isai0            = _mm_load1_pd(invsqrta+inr+0);
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vgbsum           = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166         dvdasum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_pd(ix0,jx0);
 184             dy00             = _mm_sub_pd(iy0,jy0);
 185             dz00             = _mm_sub_pd(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191
 192             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             r00              = _mm_mul_pd(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_pd(iq0,jq0);
 208             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 210
 211             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 212             isaprod          = _mm_mul_pd(isai0,isaj0);
 213             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 214             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 215
 216             /* Calculate generalized born table index - this is a separate table from the normal one,
 217              * but we use the same procedure by multiplying r with scale and truncating to integer.
 218              */
 219             rt               = _mm_mul_pd(r00,gbscale);
 220             gbitab           = _mm_cvttpd_epi32(rt);
 221             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 222             gbitab           = _mm_slli_epi32(gbitab,2);
 223
 224             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 225             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(Y,F);
 227             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 228             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(G,H);
 230             Heps             = _mm_mul_pd(gbeps,H);
 231             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 232             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 233             vgb              = _mm_mul_pd(gbqqfactor,VV);
 234
 235             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 236             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 237             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 238             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 239             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 240             velec            = _mm_mul_pd(qq00,rinv00);
 241             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 248             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             velecsum         = _mm_add_pd(velecsum,velec);
 253             vgbsum           = _mm_add_pd(vgbsum,vgb);
 254             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 255
 256             fscal            = _mm_add_pd(felec,fvdw);
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_pd(fscal,dx00);
 260             ty               = _mm_mul_pd(fscal,dy00);
 261             tz               = _mm_mul_pd(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_add_pd(fix0,tx);
 265             fiy0             = _mm_add_pd(fiy0,ty);
 266             fiz0             = _mm_add_pd(fiz0,tz);
 267
 268             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 269
 270             /* Inner loop uses 71 flops */
 271         }
 272
 273         if(jidx<j_index_end)
 274         {
 275
 276             jnrA             = jjnr[jidx];
 277             j_coord_offsetA  = DIM*jnrA;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_pd(ix0,jx0);
 285             dy00             = _mm_sub_pd(iy0,jy0);
 286             dz00             = _mm_sub_pd(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 292
 293             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = _mm_load_sd(charge+jnrA+0);
 297             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r00              = _mm_mul_pd(rsq00,rinv00);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq00             = _mm_mul_pd(iq0,jq0);
 308             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 309
 310             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 311             isaprod          = _mm_mul_pd(isai0,isaj0);
 312             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 313             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 314
 315             /* Calculate generalized born table index - this is a separate table from the normal one,
 316              * but we use the same procedure by multiplying r with scale and truncating to integer.
 317              */
 318             rt               = _mm_mul_pd(r00,gbscale);
 319             gbitab           = _mm_cvttpd_epi32(rt);
 320             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 321             gbitab           = _mm_slli_epi32(gbitab,2);
 322
 323             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 324             F                = _mm_setzero_pd();
 325             GMX_MM_TRANSPOSE2_PD(Y,F);
 326             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 327             H                = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(G,H);
 329             Heps             = _mm_mul_pd(gbeps,H);
 330             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 331             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 332             vgb              = _mm_mul_pd(gbqqfactor,VV);
 333
 334             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 335             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 336             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 337             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 338             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 339             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 340             velec            = _mm_mul_pd(qq00,rinv00);
 341             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 342
 343             /* LENNARD-JONES DISPERSION/REPULSION */
 344
 345             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 346             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 347             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 348             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 349             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 355             vgbsum           = _mm_add_pd(vgbsum,vgb);
 356             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 357             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 358
 359             fscal            = _mm_add_pd(felec,fvdw);
 360
 361             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_pd(fscal,dx00);
 365             ty               = _mm_mul_pd(fscal,dy00);
 366             tz               = _mm_mul_pd(fscal,dz00);
 367
 368             /* Update vectorial force */
 369             fix0             = _mm_add_pd(fix0,tx);
 370             fiy0             = _mm_add_pd(fiy0,ty);
 371             fiz0             = _mm_add_pd(fiz0,tz);
 372
 373             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 374
 375             /* Inner loop uses 71 flops */
 376         }
 377
 378         /* End of innermost loop */
 379
 380         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 381                                               f+i_coord_offset,fshift+i_shift_offset);
 382
 383         ggid                        = gid[iidx];
 384         /* Update potential energies */
 385         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 386         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 387         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 388         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 389         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 390
 391         /* Increment number of inner iterations */
 392         inneriter                  += j_index_end - j_index_start;
 393
 394         /* Outer loop uses 10 flops */
 395     }
 396
 397     /* Increment number of outer iterations */
 398     outeriter        += nri;
 399
 400     /* Update outer/inner flops */
 401
 402     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 403 }
 404 /*
 405  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 406  * Electrostatics interaction: GeneralizedBorn
 407  * VdW interaction:            LennardJones
 408  * Geometry:                   Particle-Particle
 409  * Calculate force/pot:        Force
 410  */
 411 void
 412 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 413                     (t_nblist                    * gmx_restrict       nlist,
 414                      rvec                        * gmx_restrict          xx,
 415                      rvec                        * gmx_restrict          ff,
 416                      t_forcerec                  * gmx_restrict          fr,
 417                      t_mdatoms                   * gmx_restrict     mdatoms,
 418                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 419                      t_nrnb                      * gmx_restrict        nrnb)
 420 {
 421     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 422      * just 0 for non-waters.
 423      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 424      * jnr indices corresponding to data put in the four positions in the SIMD register.
 425      */
 426     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 427     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 428     int              jnrA,jnrB;
 429     int              j_coord_offsetA,j_coord_offsetB;
 430     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 431     real             rcutoff_scalar;
 432     real             *shiftvec,*fshift,*x,*f;
 433     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 434     int              vdwioffset0;
 435     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B;
 437     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     __m128i          gbitab;
 442     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 443     __m128d          minushalf = _mm_set1_pd(-0.5);
 444     real             *invsqrta,*dvda,*gbtab;
 445     int              nvdwtype;
 446     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 447     int              *vdwtype;
 448     real             *vdwparam;
 449     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 450     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 451     __m128i          vfitab;
 452     __m128i          ifour       = _mm_set1_epi32(4);
 453     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 454     real             *vftab;
 455     __m128d          dummy_mask,cutoff_mask;
 456     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 457     __m128d          one     = _mm_set1_pd(1.0);
 458     __m128d          two     = _mm_set1_pd(2.0);
 459     x                = xx[0];
 460     f                = ff[0];
 461
 462     nri              = nlist->nri;
 463     iinr             = nlist->iinr;
 464     jindex           = nlist->jindex;
 465     jjnr             = nlist->jjnr;
 466     shiftidx         = nlist->shift;
 467     gid              = nlist->gid;
 468     shiftvec         = fr->shift_vec[0];
 469     fshift           = fr->fshift[0];
 470     facel            = _mm_set1_pd(fr->epsfac);
 471     charge           = mdatoms->chargeA;
 472     nvdwtype         = fr->ntype;
 473     vdwparam         = fr->nbfp;
 474     vdwtype          = mdatoms->typeA;
 475
 476     invsqrta         = fr->invsqrta;
 477     dvda             = fr->dvda;
 478     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 479     gbtab            = fr->gbtab.data;
 480     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 481
 482     /* Avoid stupid compiler warnings */
 483     jnrA = jnrB = 0;
 484     j_coord_offsetA = 0;
 485     j_coord_offsetB = 0;
 486
 487     outeriter        = 0;
 488     inneriter        = 0;
 489
 490     /* Start outer loop over neighborlists */
 491     for(iidx=0; iidx<nri; iidx++)
 492     {
 493         /* Load shift vector for this list */
 494         i_shift_offset   = DIM*shiftidx[iidx];
 495
 496         /* Load limits for loop over neighbors */
 497         j_index_start    = jindex[iidx];
 498         j_index_end      = jindex[iidx+1];
 499
 500         /* Get outer coordinate index */
 501         inr              = iinr[iidx];
 502         i_coord_offset   = DIM*inr;
 503
 504         /* Load i particle coords and add shift vector */
 505         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 506
 507         fix0             = _mm_setzero_pd();
 508         fiy0             = _mm_setzero_pd();
 509         fiz0             = _mm_setzero_pd();
 510
 511         /* Load parameters for i particles */
 512         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 513         isai0            = _mm_load1_pd(invsqrta+inr+0);
 514         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 515
 516         dvdasum          = _mm_setzero_pd();
 517
 518         /* Start inner kernel loop */
 519         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 520         {
 521
 522             /* Get j neighbor index, and coordinate index */
 523             jnrA             = jjnr[jidx];
 524             jnrB             = jjnr[jidx+1];
 525             j_coord_offsetA  = DIM*jnrA;
 526             j_coord_offsetB  = DIM*jnrB;
 527
 528             /* load j atom coordinates */
 529             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 530                                               &jx0,&jy0,&jz0);
 531
 532             /* Calculate displacement vector */
 533             dx00             = _mm_sub_pd(ix0,jx0);
 534             dy00             = _mm_sub_pd(iy0,jy0);
 535             dz00             = _mm_sub_pd(iz0,jz0);
 536
 537             /* Calculate squared distance and things based on it */
 538             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 539
 540             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 541
 542             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 543
 544             /* Load parameters for j particles */
 545             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 546             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 547             vdwjidx0A        = 2*vdwtype[jnrA+0];
 548             vdwjidx0B        = 2*vdwtype[jnrB+0];
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r00              = _mm_mul_pd(rsq00,rinv00);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq00             = _mm_mul_pd(iq0,jq0);
 558             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 559                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 560
 561             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 562             isaprod          = _mm_mul_pd(isai0,isaj0);
 563             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 564             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 565
 566             /* Calculate generalized born table index - this is a separate table from the normal one,
 567              * but we use the same procedure by multiplying r with scale and truncating to integer.
 568              */
 569             rt               = _mm_mul_pd(r00,gbscale);
 570             gbitab           = _mm_cvttpd_epi32(rt);
 571             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 572             gbitab           = _mm_slli_epi32(gbitab,2);
 573
 574             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 575             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 576             GMX_MM_TRANSPOSE2_PD(Y,F);
 577             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 578             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 579             GMX_MM_TRANSPOSE2_PD(G,H);
 580             Heps             = _mm_mul_pd(gbeps,H);
 581             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 582             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 583             vgb              = _mm_mul_pd(gbqqfactor,VV);
 584
 585             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 586             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 587             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 588             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 589             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 590             velec            = _mm_mul_pd(qq00,rinv00);
 591             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 592
 593             /* LENNARD-JONES DISPERSION/REPULSION */
 594
 595             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 596             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 597
 598             fscal            = _mm_add_pd(felec,fvdw);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_pd(fscal,dx00);
 602             ty               = _mm_mul_pd(fscal,dy00);
 603             tz               = _mm_mul_pd(fscal,dz00);
 604
 605             /* Update vectorial force */
 606             fix0             = _mm_add_pd(fix0,tx);
 607             fiy0             = _mm_add_pd(fiy0,ty);
 608             fiz0             = _mm_add_pd(fiz0,tz);
 609
 610             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 611
 612             /* Inner loop uses 64 flops */
 613         }
 614
 615         if(jidx<j_index_end)
 616         {
 617
 618             jnrA             = jjnr[jidx];
 619             j_coord_offsetA  = DIM*jnrA;
 620
 621             /* load j atom coordinates */
 622             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 623                                               &jx0,&jy0,&jz0);
 624
 625             /* Calculate displacement vector */
 626             dx00             = _mm_sub_pd(ix0,jx0);
 627             dy00             = _mm_sub_pd(iy0,jy0);
 628             dz00             = _mm_sub_pd(iz0,jz0);
 629
 630             /* Calculate squared distance and things based on it */
 631             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 632
 633             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 634
 635             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 636
 637             /* Load parameters for j particles */
 638             jq0              = _mm_load_sd(charge+jnrA+0);
 639             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 640             vdwjidx0A        = 2*vdwtype[jnrA+0];
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             r00              = _mm_mul_pd(rsq00,rinv00);
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq00             = _mm_mul_pd(iq0,jq0);
 650             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 651
 652             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 653             isaprod          = _mm_mul_pd(isai0,isaj0);
 654             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 655             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 656
 657             /* Calculate generalized born table index - this is a separate table from the normal one,
 658              * but we use the same procedure by multiplying r with scale and truncating to integer.
 659              */
 660             rt               = _mm_mul_pd(r00,gbscale);
 661             gbitab           = _mm_cvttpd_epi32(rt);
 662             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 663             gbitab           = _mm_slli_epi32(gbitab,2);
 664
 665             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 666             F                = _mm_setzero_pd();
 667             GMX_MM_TRANSPOSE2_PD(Y,F);
 668             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 669             H                = _mm_setzero_pd();
 670             GMX_MM_TRANSPOSE2_PD(G,H);
 671             Heps             = _mm_mul_pd(gbeps,H);
 672             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 673             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 674             vgb              = _mm_mul_pd(gbqqfactor,VV);
 675
 676             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 677             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 678             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 679             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 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 }