src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwCSTab_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_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_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     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_pd(fr->ic->k_rf);
 115     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_pd(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_vdw->data;
 122     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 212             rt               = _mm_mul_pd(r00,vftabscale);
 213             vfitab           = _mm_cvttpd_epi32(rt);
 214             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 215             vfitab           = _mm_slli_epi32(vfitab,3);
 216
 217             /* REACTION-FIELD ELECTROSTATICS */
 218             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 219             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 220
 221             /* CUBIC SPLINE TABLE DISPERSION */
 222             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 223             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 224             GMX_MM_TRANSPOSE2_PD(Y,F);
 225             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 226             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 227             GMX_MM_TRANSPOSE2_PD(G,H);
 228             Heps             = _mm_mul_pd(vfeps,H);
 229             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 230             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 231             vvdw6            = _mm_mul_pd(c6_00,VV);
 232             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 233             fvdw6            = _mm_mul_pd(c6_00,FF);
 234
 235             /* CUBIC SPLINE TABLE REPULSION */
 236             vfitab           = _mm_add_epi32(vfitab,ifour);
 237             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Heps             = _mm_mul_pd(vfeps,H);
 244             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 245             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 246             vvdw12           = _mm_mul_pd(c12_00,VV);
 247             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 248             fvdw12           = _mm_mul_pd(c12_00,FF);
 249             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 250             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 251
 252             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             velec            = _mm_and_pd(velec,cutoff_mask);
 256             velecsum         = _mm_add_pd(velecsum,velec);
 257             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 258             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 259
 260             fscal            = _mm_add_pd(felec,fvdw);
 261
 262             fscal            = _mm_and_pd(fscal,cutoff_mask);
 263
 264             /* Calculate temporary vectorial force */
 265             tx               = _mm_mul_pd(fscal,dx00);
 266             ty               = _mm_mul_pd(fscal,dy00);
 267             tz               = _mm_mul_pd(fscal,dz00);
 268
 269             /* Update vectorial force */
 270             fix0             = _mm_add_pd(fix0,tx);
 271             fiy0             = _mm_add_pd(fiy0,ty);
 272             fiz0             = _mm_add_pd(fiz0,tz);
 273
 274             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 275
 276             }
 277
 278             /* Inner loop uses 72 flops */
 279         }
 280
 281         if(jidx<j_index_end)
 282         {
 283
 284             jnrA             = jjnr[jidx];
 285             j_coord_offsetA  = DIM*jnrA;
 286
 287             /* load j atom coordinates */
 288             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 289                                               &jx0,&jy0,&jz0);
 290
 291             /* Calculate displacement vector */
 292             dx00             = _mm_sub_pd(ix0,jx0);
 293             dy00             = _mm_sub_pd(iy0,jy0);
 294             dz00             = _mm_sub_pd(iz0,jz0);
 295
 296             /* Calculate squared distance and things based on it */
 297             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 298
 299             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 300
 301             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 302
 303             /* Load parameters for j particles */
 304             jq0              = _mm_load_sd(charge+jnrA+0);
 305             vdwjidx0A        = 2*vdwtype[jnrA+0];
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq00,rcutoff2))
 312             {
 313
 314             r00              = _mm_mul_pd(rsq00,rinv00);
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq00             = _mm_mul_pd(iq0,jq0);
 318             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 319
 320             /* Calculate table index by multiplying r with table scale and truncate to integer */
 321             rt               = _mm_mul_pd(r00,vftabscale);
 322             vfitab           = _mm_cvttpd_epi32(rt);
 323             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 324             vfitab           = _mm_slli_epi32(vfitab,3);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 328             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 329
 330             /* CUBIC SPLINE TABLE DISPERSION */
 331             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 332             F                = _mm_setzero_pd();
 333             GMX_MM_TRANSPOSE2_PD(Y,F);
 334             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 335             H                = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(G,H);
 337             Heps             = _mm_mul_pd(vfeps,H);
 338             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 339             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 340             vvdw6            = _mm_mul_pd(c6_00,VV);
 341             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 342             fvdw6            = _mm_mul_pd(c6_00,FF);
 343
 344             /* CUBIC SPLINE TABLE REPULSION */
 345             vfitab           = _mm_add_epi32(vfitab,ifour);
 346             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 347             F                = _mm_setzero_pd();
 348             GMX_MM_TRANSPOSE2_PD(Y,F);
 349             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 350             H                = _mm_setzero_pd();
 351             GMX_MM_TRANSPOSE2_PD(G,H);
 352             Heps             = _mm_mul_pd(vfeps,H);
 353             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 354             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 355             vvdw12           = _mm_mul_pd(c12_00,VV);
 356             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 357             fvdw12           = _mm_mul_pd(c12_00,FF);
 358             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 359             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 360
 361             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_and_pd(velec,cutoff_mask);
 365             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 368             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 369             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 370
 371             fscal            = _mm_add_pd(felec,fvdw);
 372
 373             fscal            = _mm_and_pd(fscal,cutoff_mask);
 374
 375             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_pd(fscal,dx00);
 379             ty               = _mm_mul_pd(fscal,dy00);
 380             tz               = _mm_mul_pd(fscal,dz00);
 381
 382             /* Update vectorial force */
 383             fix0             = _mm_add_pd(fix0,tx);
 384             fiy0             = _mm_add_pd(fiy0,ty);
 385             fiz0             = _mm_add_pd(fiz0,tz);
 386
 387             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 388
 389             }
 390
 391             /* Inner loop uses 72 flops */
 392         }
 393
 394         /* End of innermost loop */
 395
 396         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 397                                               f+i_coord_offset,fshift+i_shift_offset);
 398
 399         ggid                        = gid[iidx];
 400         /* Update potential energies */
 401         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 402         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 403
 404         /* Increment number of inner iterations */
 405         inneriter                  += j_index_end - j_index_start;
 406
 407         /* Outer loop uses 9 flops */
 408     }
 409
 410     /* Increment number of outer iterations */
 411     outeriter        += nri;
 412
 413     /* Update outer/inner flops */
 414
 415     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*72);
 416 }
 417 /*
 418  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_double
 419  * Electrostatics interaction: ReactionField
 420  * VdW interaction:            CubicSplineTable
 421  * Geometry:                   Particle-Particle
 422  * Calculate force/pot:        Force
 423  */
 424 void
 425 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_double
 426                     (t_nblist                    * gmx_restrict       nlist,
 427                      rvec                        * gmx_restrict          xx,
 428                      rvec                        * gmx_restrict          ff,
 429                      t_forcerec                  * gmx_restrict          fr,
 430                      t_mdatoms                   * gmx_restrict     mdatoms,
 431                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 432                      t_nrnb                      * gmx_restrict        nrnb)
 433 {
 434     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 435      * just 0 for non-waters.
 436      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 437      * jnr indices corresponding to data put in the four positions in the SIMD register.
 438      */
 439     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 440     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 441     int              jnrA,jnrB;
 442     int              j_coord_offsetA,j_coord_offsetB;
 443     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 444     real             rcutoff_scalar;
 445     real             *shiftvec,*fshift,*x,*f;
 446     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 447     int              vdwioffset0;
 448     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 449     int              vdwjidx0A,vdwjidx0B;
 450     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 451     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 452     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 453     real             *charge;
 454     int              nvdwtype;
 455     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 456     int              *vdwtype;
 457     real             *vdwparam;
 458     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 459     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 460     __m128i          vfitab;
 461     __m128i          ifour       = _mm_set1_epi32(4);
 462     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 463     real             *vftab;
 464     __m128d          dummy_mask,cutoff_mask;
 465     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 466     __m128d          one     = _mm_set1_pd(1.0);
 467     __m128d          two     = _mm_set1_pd(2.0);
 468     x                = xx[0];
 469     f                = ff[0];
 470
 471     nri              = nlist->nri;
 472     iinr             = nlist->iinr;
 473     jindex           = nlist->jindex;
 474     jjnr             = nlist->jjnr;
 475     shiftidx         = nlist->shift;
 476     gid              = nlist->gid;
 477     shiftvec         = fr->shift_vec[0];
 478     fshift           = fr->fshift[0];
 479     facel            = _mm_set1_pd(fr->epsfac);
 480     charge           = mdatoms->chargeA;
 481     krf              = _mm_set1_pd(fr->ic->k_rf);
 482     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 483     crf              = _mm_set1_pd(fr->ic->c_rf);
 484     nvdwtype         = fr->ntype;
 485     vdwparam         = fr->nbfp;
 486     vdwtype          = mdatoms->typeA;
 487
 488     vftab            = kernel_data->table_vdw->data;
 489     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 490
 491     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 492     rcutoff_scalar   = fr->rcoulomb;
 493     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 494     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 495
 496     /* Avoid stupid compiler warnings */
 497     jnrA = jnrB = 0;
 498     j_coord_offsetA = 0;
 499     j_coord_offsetB = 0;
 500
 501     outeriter        = 0;
 502     inneriter        = 0;
 503
 504     /* Start outer loop over neighborlists */
 505     for(iidx=0; iidx<nri; iidx++)
 506     {
 507         /* Load shift vector for this list */
 508         i_shift_offset   = DIM*shiftidx[iidx];
 509
 510         /* Load limits for loop over neighbors */
 511         j_index_start    = jindex[iidx];
 512         j_index_end      = jindex[iidx+1];
 513
 514         /* Get outer coordinate index */
 515         inr              = iinr[iidx];
 516         i_coord_offset   = DIM*inr;
 517
 518         /* Load i particle coords and add shift vector */
 519         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 520
 521         fix0             = _mm_setzero_pd();
 522         fiy0             = _mm_setzero_pd();
 523         fiz0             = _mm_setzero_pd();
 524
 525         /* Load parameters for i particles */
 526         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 527         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 528
 529         /* Start inner kernel loop */
 530         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 531         {
 532
 533             /* Get j neighbor index, and coordinate index */
 534             jnrA             = jjnr[jidx];
 535             jnrB             = jjnr[jidx+1];
 536             j_coord_offsetA  = DIM*jnrA;
 537             j_coord_offsetB  = DIM*jnrB;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 541                                               &jx0,&jy0,&jz0);
 542
 543             /* Calculate displacement vector */
 544             dx00             = _mm_sub_pd(ix0,jx0);
 545             dy00             = _mm_sub_pd(iy0,jy0);
 546             dz00             = _mm_sub_pd(iz0,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 550
 551             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 552
 553             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 554
 555             /* Load parameters for j particles */
 556             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558             vdwjidx0B        = 2*vdwtype[jnrB+0];
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm_any_lt(rsq00,rcutoff2))
 565             {
 566
 567             r00              = _mm_mul_pd(rsq00,rinv00);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq00             = _mm_mul_pd(iq0,jq0);
 571             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 572                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 573
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm_mul_pd(r00,vftabscale);
 576             vfitab           = _mm_cvttpd_epi32(rt);
 577             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 578             vfitab           = _mm_slli_epi32(vfitab,3);
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 582
 583             /* CUBIC SPLINE TABLE DISPERSION */
 584             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 585             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 586             GMX_MM_TRANSPOSE2_PD(Y,F);
 587             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 588             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 589             GMX_MM_TRANSPOSE2_PD(G,H);
 590             Heps             = _mm_mul_pd(vfeps,H);
 591             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 592             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 593             fvdw6            = _mm_mul_pd(c6_00,FF);
 594
 595             /* CUBIC SPLINE TABLE REPULSION */
 596             vfitab           = _mm_add_epi32(vfitab,ifour);
 597             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 598             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 599             GMX_MM_TRANSPOSE2_PD(Y,F);
 600             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 601             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 602             GMX_MM_TRANSPOSE2_PD(G,H);
 603             Heps             = _mm_mul_pd(vfeps,H);
 604             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 605             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 606             fvdw12           = _mm_mul_pd(c12_00,FF);
 607             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 608
 609             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 610
 611             fscal            = _mm_add_pd(felec,fvdw);
 612
 613             fscal            = _mm_and_pd(fscal,cutoff_mask);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx00);
 617             ty               = _mm_mul_pd(fscal,dy00);
 618             tz               = _mm_mul_pd(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_add_pd(fix0,tx);
 622             fiy0             = _mm_add_pd(fiy0,ty);
 623             fiz0             = _mm_add_pd(fiz0,tz);
 624
 625             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 626
 627             }
 628
 629             /* Inner loop uses 57 flops */
 630         }
 631
 632         if(jidx<j_index_end)
 633         {
 634
 635             jnrA             = jjnr[jidx];
 636             j_coord_offsetA  = DIM*jnrA;
 637
 638             /* load j atom coordinates */
 639             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 640                                               &jx0,&jy0,&jz0);
 641
 642             /* Calculate displacement vector */
 643             dx00             = _mm_sub_pd(ix0,jx0);
 644             dy00             = _mm_sub_pd(iy0,jy0);
 645             dz00             = _mm_sub_pd(iz0,jz0);
 646
 647             /* Calculate squared distance and things based on it */
 648             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 649
 650             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 651
 652             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 653
 654             /* Load parameters for j particles */
 655             jq0              = _mm_load_sd(charge+jnrA+0);
 656             vdwjidx0A        = 2*vdwtype[jnrA+0];
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             if (gmx_mm_any_lt(rsq00,rcutoff2))
 663             {
 664
 665             r00              = _mm_mul_pd(rsq00,rinv00);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm_mul_pd(iq0,jq0);
 669             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 670
 671             /* Calculate table index by multiplying r with table scale and truncate to integer */
 672             rt               = _mm_mul_pd(r00,vftabscale);
 673             vfitab           = _mm_cvttpd_epi32(rt);
 674             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 675             vfitab           = _mm_slli_epi32(vfitab,3);
 676
 677             /* REACTION-FIELD ELECTROSTATICS */
 678             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 679
 680             /* CUBIC SPLINE TABLE DISPERSION */
 681             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 682             F                = _mm_setzero_pd();
 683             GMX_MM_TRANSPOSE2_PD(Y,F);
 684             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 685             H                = _mm_setzero_pd();
 686             GMX_MM_TRANSPOSE2_PD(G,H);
 687             Heps             = _mm_mul_pd(vfeps,H);
 688             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 689             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 690             fvdw6            = _mm_mul_pd(c6_00,FF);
 691
 692             /* CUBIC SPLINE TABLE REPULSION */
 693             vfitab           = _mm_add_epi32(vfitab,ifour);
 694             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 695             F                = _mm_setzero_pd();
 696             GMX_MM_TRANSPOSE2_PD(Y,F);
 697             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 698             H                = _mm_setzero_pd();
 699             GMX_MM_TRANSPOSE2_PD(G,H);
 700             Heps             = _mm_mul_pd(vfeps,H);
 701             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 702             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 703             fvdw12           = _mm_mul_pd(c12_00,FF);
 704             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 705
 706             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 707
 708             fscal            = _mm_add_pd(felec,fvdw);
 709
 710             fscal            = _mm_and_pd(fscal,cutoff_mask);
 711
 712             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm_mul_pd(fscal,dx00);
 716             ty               = _mm_mul_pd(fscal,dy00);
 717             tz               = _mm_mul_pd(fscal,dz00);
 718
 719             /* Update vectorial force */
 720             fix0             = _mm_add_pd(fix0,tx);
 721             fiy0             = _mm_add_pd(fiy0,ty);
 722             fiz0             = _mm_add_pd(fiz0,tz);
 723
 724             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 725
 726             }
 727
 728             /* Inner loop uses 57 flops */
 729         }
 730
 731         /* End of innermost loop */
 732
 733         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 734                                               f+i_coord_offset,fshift+i_shift_offset);
 735
 736         /* Increment number of inner iterations */
 737         inneriter                  += j_index_end - j_index_start;
 738
 739         /* Outer loop uses 7 flops */
 740     }
 741
 742     /* Increment number of outer iterations */
 743     outeriter        += nri;
 744
 745     /* Update outer/inner flops */
 746
 747     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 748 }