src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwCSTab_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_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_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     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm_set1_pd(fr->ic->k_rf);
 117     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_pd(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_pd();
 157         fiy0             = _mm_setzero_pd();
 158         fiz0             = _mm_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 214             rt               = _mm_mul_pd(r00,vftabscale);
 215             vfitab           = _mm_cvttpd_epi32(rt);
 216             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 217             vfitab           = _mm_slli_epi32(vfitab,3);
 218
 219             /* REACTION-FIELD ELECTROSTATICS */
 220             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 221             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 222
 223             /* CUBIC SPLINE TABLE DISPERSION */
 224             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 225             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(Y,F);
 227             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 228             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(G,H);
 230             Heps             = _mm_mul_pd(vfeps,H);
 231             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 232             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 233             vvdw6            = _mm_mul_pd(c6_00,VV);
 234             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 235             fvdw6            = _mm_mul_pd(c6_00,FF);
 236
 237             /* CUBIC SPLINE TABLE REPULSION */
 238             vfitab           = _mm_add_epi32(vfitab,ifour);
 239             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 240             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 241             GMX_MM_TRANSPOSE2_PD(Y,F);
 242             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 243             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 244             GMX_MM_TRANSPOSE2_PD(G,H);
 245             Heps             = _mm_mul_pd(vfeps,H);
 246             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 247             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 248             vvdw12           = _mm_mul_pd(c12_00,VV);
 249             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 250             fvdw12           = _mm_mul_pd(c12_00,FF);
 251             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 252             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 253
 254             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velec            = _mm_and_pd(velec,cutoff_mask);
 258             velecsum         = _mm_add_pd(velecsum,velec);
 259             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 260             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 261
 262             fscal            = _mm_add_pd(felec,fvdw);
 263
 264             fscal            = _mm_and_pd(fscal,cutoff_mask);
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_pd(fscal,dx00);
 268             ty               = _mm_mul_pd(fscal,dy00);
 269             tz               = _mm_mul_pd(fscal,dz00);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_add_pd(fix0,tx);
 273             fiy0             = _mm_add_pd(fiy0,ty);
 274             fiz0             = _mm_add_pd(fiz0,tz);
 275
 276             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 277
 278             }
 279
 280             /* Inner loop uses 72 flops */
 281         }
 282
 283         if(jidx<j_index_end)
 284         {
 285
 286             jnrA             = jjnr[jidx];
 287             j_coord_offsetA  = DIM*jnrA;
 288
 289             /* load j atom coordinates */
 290             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 291                                               &jx0,&jy0,&jz0);
 292
 293             /* Calculate displacement vector */
 294             dx00             = _mm_sub_pd(ix0,jx0);
 295             dy00             = _mm_sub_pd(iy0,jy0);
 296             dz00             = _mm_sub_pd(iz0,jz0);
 297
 298             /* Calculate squared distance and things based on it */
 299             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 300
 301             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 302
 303             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 304
 305             /* Load parameters for j particles */
 306             jq0              = _mm_load_sd(charge+jnrA+0);
 307             vdwjidx0A        = 2*vdwtype[jnrA+0];
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             if (gmx_mm_any_lt(rsq00,rcutoff2))
 314             {
 315
 316             r00              = _mm_mul_pd(rsq00,rinv00);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq00             = _mm_mul_pd(iq0,jq0);
 320             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 321
 322             /* Calculate table index by multiplying r with table scale and truncate to integer */
 323             rt               = _mm_mul_pd(r00,vftabscale);
 324             vfitab           = _mm_cvttpd_epi32(rt);
 325             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 326             vfitab           = _mm_slli_epi32(vfitab,3);
 327
 328             /* REACTION-FIELD ELECTROSTATICS */
 329             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 330             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 331
 332             /* CUBIC SPLINE TABLE DISPERSION */
 333             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 334             F                = _mm_setzero_pd();
 335             GMX_MM_TRANSPOSE2_PD(Y,F);
 336             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 337             H                = _mm_setzero_pd();
 338             GMX_MM_TRANSPOSE2_PD(G,H);
 339             Heps             = _mm_mul_pd(vfeps,H);
 340             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 341             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 342             vvdw6            = _mm_mul_pd(c6_00,VV);
 343             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 344             fvdw6            = _mm_mul_pd(c6_00,FF);
 345
 346             /* CUBIC SPLINE TABLE REPULSION */
 347             vfitab           = _mm_add_epi32(vfitab,ifour);
 348             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 349             F                = _mm_setzero_pd();
 350             GMX_MM_TRANSPOSE2_PD(Y,F);
 351             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 352             H                = _mm_setzero_pd();
 353             GMX_MM_TRANSPOSE2_PD(G,H);
 354             Heps             = _mm_mul_pd(vfeps,H);
 355             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 356             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 357             vvdw12           = _mm_mul_pd(c12_00,VV);
 358             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 359             fvdw12           = _mm_mul_pd(c12_00,FF);
 360             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 361             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 362
 363             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velec            = _mm_and_pd(velec,cutoff_mask);
 367             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 368             velecsum         = _mm_add_pd(velecsum,velec);
 369             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 370             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 371             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 372
 373             fscal            = _mm_add_pd(felec,fvdw);
 374
 375             fscal            = _mm_and_pd(fscal,cutoff_mask);
 376
 377             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_pd(fscal,dx00);
 381             ty               = _mm_mul_pd(fscal,dy00);
 382             tz               = _mm_mul_pd(fscal,dz00);
 383
 384             /* Update vectorial force */
 385             fix0             = _mm_add_pd(fix0,tx);
 386             fiy0             = _mm_add_pd(fiy0,ty);
 387             fiz0             = _mm_add_pd(fiz0,tz);
 388
 389             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 390
 391             }
 392
 393             /* Inner loop uses 72 flops */
 394         }
 395
 396         /* End of innermost loop */
 397
 398         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 399                                               f+i_coord_offset,fshift+i_shift_offset);
 400
 401         ggid                        = gid[iidx];
 402         /* Update potential energies */
 403         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 404         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 405
 406         /* Increment number of inner iterations */
 407         inneriter                  += j_index_end - j_index_start;
 408
 409         /* Outer loop uses 9 flops */
 410     }
 411
 412     /* Increment number of outer iterations */
 413     outeriter        += nri;
 414
 415     /* Update outer/inner flops */
 416
 417     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*72);
 418 }
 419 /*
 420  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_double
 421  * Electrostatics interaction: ReactionField
 422  * VdW interaction:            CubicSplineTable
 423  * Geometry:                   Particle-Particle
 424  * Calculate force/pot:        Force
 425  */
 426 void
 427 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_double
 428                     (t_nblist                    * gmx_restrict       nlist,
 429                      rvec                        * gmx_restrict          xx,
 430                      rvec                        * gmx_restrict          ff,
 431                      t_forcerec                  * gmx_restrict          fr,
 432                      t_mdatoms                   * gmx_restrict     mdatoms,
 433                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 434                      t_nrnb                      * gmx_restrict        nrnb)
 435 {
 436     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 437      * just 0 for non-waters.
 438      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 439      * jnr indices corresponding to data put in the four positions in the SIMD register.
 440      */
 441     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 442     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 443     int              jnrA,jnrB;
 444     int              j_coord_offsetA,j_coord_offsetB;
 445     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 446     real             rcutoff_scalar;
 447     real             *shiftvec,*fshift,*x,*f;
 448     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 449     int              vdwioffset0;
 450     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 451     int              vdwjidx0A,vdwjidx0B;
 452     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 453     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 454     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 455     real             *charge;
 456     int              nvdwtype;
 457     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 458     int              *vdwtype;
 459     real             *vdwparam;
 460     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 461     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 462     __m128i          vfitab;
 463     __m128i          ifour       = _mm_set1_epi32(4);
 464     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 465     real             *vftab;
 466     __m128d          dummy_mask,cutoff_mask;
 467     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 468     __m128d          one     = _mm_set1_pd(1.0);
 469     __m128d          two     = _mm_set1_pd(2.0);
 470     x                = xx[0];
 471     f                = ff[0];
 472
 473     nri              = nlist->nri;
 474     iinr             = nlist->iinr;
 475     jindex           = nlist->jindex;
 476     jjnr             = nlist->jjnr;
 477     shiftidx         = nlist->shift;
 478     gid              = nlist->gid;
 479     shiftvec         = fr->shift_vec[0];
 480     fshift           = fr->fshift[0];
 481     facel            = _mm_set1_pd(fr->epsfac);
 482     charge           = mdatoms->chargeA;
 483     krf              = _mm_set1_pd(fr->ic->k_rf);
 484     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 485     crf              = _mm_set1_pd(fr->ic->c_rf);
 486     nvdwtype         = fr->ntype;
 487     vdwparam         = fr->nbfp;
 488     vdwtype          = mdatoms->typeA;
 489
 490     vftab            = kernel_data->table_vdw->data;
 491     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 492
 493     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 494     rcutoff_scalar   = fr->rcoulomb;
 495     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 496     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 497
 498     /* Avoid stupid compiler warnings */
 499     jnrA = jnrB = 0;
 500     j_coord_offsetA = 0;
 501     j_coord_offsetB = 0;
 502
 503     outeriter        = 0;
 504     inneriter        = 0;
 505
 506     /* Start outer loop over neighborlists */
 507     for(iidx=0; iidx<nri; iidx++)
 508     {
 509         /* Load shift vector for this list */
 510         i_shift_offset   = DIM*shiftidx[iidx];
 511
 512         /* Load limits for loop over neighbors */
 513         j_index_start    = jindex[iidx];
 514         j_index_end      = jindex[iidx+1];
 515
 516         /* Get outer coordinate index */
 517         inr              = iinr[iidx];
 518         i_coord_offset   = DIM*inr;
 519
 520         /* Load i particle coords and add shift vector */
 521         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 522
 523         fix0             = _mm_setzero_pd();
 524         fiy0             = _mm_setzero_pd();
 525         fiz0             = _mm_setzero_pd();
 526
 527         /* Load parameters for i particles */
 528         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 529         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 530
 531         /* Start inner kernel loop */
 532         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 533         {
 534
 535             /* Get j neighbor index, and coordinate index */
 536             jnrA             = jjnr[jidx];
 537             jnrB             = jjnr[jidx+1];
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 543                                               &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm_sub_pd(ix0,jx0);
 547             dy00             = _mm_sub_pd(iy0,jy0);
 548             dz00             = _mm_sub_pd(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 552
 553             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 554
 555             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 556
 557             /* Load parameters for j particles */
 558             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 559             vdwjidx0A        = 2*vdwtype[jnrA+0];
 560             vdwjidx0B        = 2*vdwtype[jnrB+0];
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm_any_lt(rsq00,rcutoff2))
 567             {
 568
 569             r00              = _mm_mul_pd(rsq00,rinv00);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq00             = _mm_mul_pd(iq0,jq0);
 573             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 574                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm_mul_pd(r00,vftabscale);
 578             vfitab           = _mm_cvttpd_epi32(rt);
 579             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 580             vfitab           = _mm_slli_epi32(vfitab,3);
 581
 582             /* REACTION-FIELD ELECTROSTATICS */
 583             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 584
 585             /* CUBIC SPLINE TABLE DISPERSION */
 586             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 587             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 588             GMX_MM_TRANSPOSE2_PD(Y,F);
 589             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 590             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 591             GMX_MM_TRANSPOSE2_PD(G,H);
 592             Heps             = _mm_mul_pd(vfeps,H);
 593             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 594             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 595             fvdw6            = _mm_mul_pd(c6_00,FF);
 596
 597             /* CUBIC SPLINE TABLE REPULSION */
 598             vfitab           = _mm_add_epi32(vfitab,ifour);
 599             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 600             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 601             GMX_MM_TRANSPOSE2_PD(Y,F);
 602             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 603             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 604             GMX_MM_TRANSPOSE2_PD(G,H);
 605             Heps             = _mm_mul_pd(vfeps,H);
 606             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 607             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 608             fvdw12           = _mm_mul_pd(c12_00,FF);
 609             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 610
 611             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 612
 613             fscal            = _mm_add_pd(felec,fvdw);
 614
 615             fscal            = _mm_and_pd(fscal,cutoff_mask);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_pd(fscal,dx00);
 619             ty               = _mm_mul_pd(fscal,dy00);
 620             tz               = _mm_mul_pd(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_add_pd(fix0,tx);
 624             fiy0             = _mm_add_pd(fiy0,ty);
 625             fiz0             = _mm_add_pd(fiz0,tz);
 626
 627             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 628
 629             }
 630
 631             /* Inner loop uses 57 flops */
 632         }
 633
 634         if(jidx<j_index_end)
 635         {
 636
 637             jnrA             = jjnr[jidx];
 638             j_coord_offsetA  = DIM*jnrA;
 639
 640             /* load j atom coordinates */
 641             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 642                                               &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm_sub_pd(ix0,jx0);
 646             dy00             = _mm_sub_pd(iy0,jy0);
 647             dz00             = _mm_sub_pd(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 651
 652             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 653
 654             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 655
 656             /* Load parameters for j particles */
 657             jq0              = _mm_load_sd(charge+jnrA+0);
 658             vdwjidx0A        = 2*vdwtype[jnrA+0];
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm_any_lt(rsq00,rcutoff2))
 665             {
 666
 667             r00              = _mm_mul_pd(rsq00,rinv00);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq00             = _mm_mul_pd(iq0,jq0);
 671             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 672
 673             /* Calculate table index by multiplying r with table scale and truncate to integer */
 674             rt               = _mm_mul_pd(r00,vftabscale);
 675             vfitab           = _mm_cvttpd_epi32(rt);
 676             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 677             vfitab           = _mm_slli_epi32(vfitab,3);
 678
 679             /* REACTION-FIELD ELECTROSTATICS */
 680             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 681
 682             /* CUBIC SPLINE TABLE DISPERSION */
 683             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 684             F                = _mm_setzero_pd();
 685             GMX_MM_TRANSPOSE2_PD(Y,F);
 686             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 687             H                = _mm_setzero_pd();
 688             GMX_MM_TRANSPOSE2_PD(G,H);
 689             Heps             = _mm_mul_pd(vfeps,H);
 690             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 691             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 692             fvdw6            = _mm_mul_pd(c6_00,FF);
 693
 694             /* CUBIC SPLINE TABLE REPULSION */
 695             vfitab           = _mm_add_epi32(vfitab,ifour);
 696             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 697             F                = _mm_setzero_pd();
 698             GMX_MM_TRANSPOSE2_PD(Y,F);
 699             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 700             H                = _mm_setzero_pd();
 701             GMX_MM_TRANSPOSE2_PD(G,H);
 702             Heps             = _mm_mul_pd(vfeps,H);
 703             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 704             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 705             fvdw12           = _mm_mul_pd(c12_00,FF);
 706             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 707
 708             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 709
 710             fscal            = _mm_add_pd(felec,fvdw);
 711
 712             fscal            = _mm_and_pd(fscal,cutoff_mask);
 713
 714             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 715
 716             /* Calculate temporary vectorial force */
 717             tx               = _mm_mul_pd(fscal,dx00);
 718             ty               = _mm_mul_pd(fscal,dy00);
 719             tz               = _mm_mul_pd(fscal,dz00);
 720
 721             /* Update vectorial force */
 722             fix0             = _mm_add_pd(fix0,tx);
 723             fiy0             = _mm_add_pd(fiy0,ty);
 724             fiz0             = _mm_add_pd(fiz0,tz);
 725
 726             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 727
 728             }
 729
 730             /* Inner loop uses 57 flops */
 731         }
 732
 733         /* End of innermost loop */
 734
 735         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 736                                               f+i_coord_offset,fshift+i_shift_offset);
 737
 738         /* Increment number of inner iterations */
 739         inneriter                  += j_index_end - j_index_start;
 740
 741         /* Outer loop uses 7 flops */
 742     }
 743
 744     /* Increment number of outer iterations */
 745     outeriter        += nri;
 746
 747     /* Update outer/inner flops */
 748
 749     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 750 }