src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_128_fma_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,twovfeps;
  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     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_pd();
 149         fiy0             = _mm_setzero_pd();
 150         fiz0             = _mm_setzero_pd();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 183
 184             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             r00              = _mm_mul_pd(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm_mul_pd(iq0,jq0);
 199             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 200                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 201
 202             /* Calculate table index by multiplying r with table scale and truncate to integer */
 203             rt               = _mm_mul_pd(r00,vftabscale);
 204             vfitab           = _mm_cvttpd_epi32(rt);
 205 #ifdef __XOP__
 206             vfeps            = _mm_frcz_pd(rt);
 207 #else
 208             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 209 #endif
 210             twovfeps         = _mm_add_pd(vfeps,vfeps);
 211             vfitab           = _mm_slli_epi32(vfitab,3);
 212
 213             /* COULOMB ELECTROSTATICS */
 214             velec            = _mm_mul_pd(qq00,rinv00);
 215             felec            = _mm_mul_pd(velec,rinvsq00);
 216
 217             /* CUBIC SPLINE TABLE DISPERSION */
 218             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 219             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 220             GMX_MM_TRANSPOSE2_PD(Y,F);
 221             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 222             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 223             GMX_MM_TRANSPOSE2_PD(G,H);
 224             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 225             VV               = _mm_macc_pd(vfeps,Fp,Y);
 226             vvdw6            = _mm_mul_pd(c6_00,VV);
 227             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 228             fvdw6            = _mm_mul_pd(c6_00,FF);
 229
 230             /* CUBIC SPLINE TABLE REPULSION */
 231             vfitab           = _mm_add_epi32(vfitab,ifour);
 232             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 233             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 234             GMX_MM_TRANSPOSE2_PD(Y,F);
 235             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 236             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 237             GMX_MM_TRANSPOSE2_PD(G,H);
 238             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 239             VV               = _mm_macc_pd(vfeps,Fp,Y);
 240             vvdw12           = _mm_mul_pd(c12_00,VV);
 241             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 242             fvdw12           = _mm_mul_pd(c12_00,FF);
 243             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 244             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 249
 250             fscal            = _mm_add_pd(felec,fvdw);
 251
 252             /* Update vectorial force */
 253             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 254             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 255             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 256
 257             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 258                                                    _mm_mul_pd(dx00,fscal),
 259                                                    _mm_mul_pd(dy00,fscal),
 260                                                    _mm_mul_pd(dz00,fscal));
 261
 262             /* Inner loop uses 66 flops */
 263         }
 264
 265         if(jidx<j_index_end)
 266         {
 267
 268             jnrA             = jjnr[jidx];
 269             j_coord_offsetA  = DIM*jnrA;
 270
 271             /* load j atom coordinates */
 272             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 273                                               &jx0,&jy0,&jz0);
 274
 275             /* Calculate displacement vector */
 276             dx00             = _mm_sub_pd(ix0,jx0);
 277             dy00             = _mm_sub_pd(iy0,jy0);
 278             dz00             = _mm_sub_pd(iz0,jz0);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 282
 283             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 284
 285             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 286
 287             /* Load parameters for j particles */
 288             jq0              = _mm_load_sd(charge+jnrA+0);
 289             vdwjidx0A        = 2*vdwtype[jnrA+0];
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             r00              = _mm_mul_pd(rsq00,rinv00);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm_mul_pd(iq0,jq0);
 299             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 300
 301             /* Calculate table index by multiplying r with table scale and truncate to integer */
 302             rt               = _mm_mul_pd(r00,vftabscale);
 303             vfitab           = _mm_cvttpd_epi32(rt);
 304 #ifdef __XOP__
 305             vfeps            = _mm_frcz_pd(rt);
 306 #else
 307             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 308 #endif
 309             twovfeps         = _mm_add_pd(vfeps,vfeps);
 310             vfitab           = _mm_slli_epi32(vfitab,3);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm_mul_pd(qq00,rinv00);
 314             felec            = _mm_mul_pd(velec,rinvsq00);
 315
 316             /* CUBIC SPLINE TABLE DISPERSION */
 317             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 318             F                = _mm_setzero_pd();
 319             GMX_MM_TRANSPOSE2_PD(Y,F);
 320             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 321             H                = _mm_setzero_pd();
 322             GMX_MM_TRANSPOSE2_PD(G,H);
 323             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 324             VV               = _mm_macc_pd(vfeps,Fp,Y);
 325             vvdw6            = _mm_mul_pd(c6_00,VV);
 326             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 327             fvdw6            = _mm_mul_pd(c6_00,FF);
 328
 329             /* CUBIC SPLINE TABLE REPULSION */
 330             vfitab           = _mm_add_epi32(vfitab,ifour);
 331             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 332             F                = _mm_setzero_pd();
 333             GMX_MM_TRANSPOSE2_PD(Y,F);
 334             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 335             H                = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(G,H);
 337             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 338             VV               = _mm_macc_pd(vfeps,Fp,Y);
 339             vvdw12           = _mm_mul_pd(c12_00,VV);
 340             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 341             fvdw12           = _mm_mul_pd(c12_00,FF);
 342             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 343             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 349             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_pd(felec,fvdw);
 352
 353             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 354
 355             /* Update vectorial force */
 356             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 357             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 358             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 359
 360             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 361                                                    _mm_mul_pd(dx00,fscal),
 362                                                    _mm_mul_pd(dy00,fscal),
 363                                                    _mm_mul_pd(dz00,fscal));
 364
 365             /* Inner loop uses 66 flops */
 366         }
 367
 368         /* End of innermost loop */
 369
 370         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 371                                               f+i_coord_offset,fshift+i_shift_offset);
 372
 373         ggid                        = gid[iidx];
 374         /* Update potential energies */
 375         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 376         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 377
 378         /* Increment number of inner iterations */
 379         inneriter                  += j_index_end - j_index_start;
 380
 381         /* Outer loop uses 9 flops */
 382     }
 383
 384     /* Increment number of outer iterations */
 385     outeriter        += nri;
 386
 387     /* Update outer/inner flops */
 388
 389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*66);
 390 }
 391 /*
 392  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 393  * Electrostatics interaction: Coulomb
 394  * VdW interaction:            CubicSplineTable
 395  * Geometry:                   Particle-Particle
 396  * Calculate force/pot:        Force
 397  */
 398 void
 399 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 400                     (t_nblist                    * gmx_restrict       nlist,
 401                      rvec                        * gmx_restrict          xx,
 402                      rvec                        * gmx_restrict          ff,
 403                      t_forcerec                  * gmx_restrict          fr,
 404                      t_mdatoms                   * gmx_restrict     mdatoms,
 405                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 406                      t_nrnb                      * gmx_restrict        nrnb)
 407 {
 408     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 409      * just 0 for non-waters.
 410      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 411      * jnr indices corresponding to data put in the four positions in the SIMD register.
 412      */
 413     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 414     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 415     int              jnrA,jnrB;
 416     int              j_coord_offsetA,j_coord_offsetB;
 417     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 418     real             rcutoff_scalar;
 419     real             *shiftvec,*fshift,*x,*f;
 420     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 421     int              vdwioffset0;
 422     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 423     int              vdwjidx0A,vdwjidx0B;
 424     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 425     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 426     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 427     real             *charge;
 428     int              nvdwtype;
 429     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 430     int              *vdwtype;
 431     real             *vdwparam;
 432     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 433     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 434     __m128i          vfitab;
 435     __m128i          ifour       = _mm_set1_epi32(4);
 436     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 437     real             *vftab;
 438     __m128d          dummy_mask,cutoff_mask;
 439     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 440     __m128d          one     = _mm_set1_pd(1.0);
 441     __m128d          two     = _mm_set1_pd(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     facel            = _mm_set1_pd(fr->epsfac);
 454     charge           = mdatoms->chargeA;
 455     nvdwtype         = fr->ntype;
 456     vdwparam         = fr->nbfp;
 457     vdwtype          = mdatoms->typeA;
 458
 459     vftab            = kernel_data->table_vdw->data;
 460     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 461
 462     /* Avoid stupid compiler warnings */
 463     jnrA = jnrB = 0;
 464     j_coord_offsetA = 0;
 465     j_coord_offsetB = 0;
 466
 467     outeriter        = 0;
 468     inneriter        = 0;
 469
 470     /* Start outer loop over neighborlists */
 471     for(iidx=0; iidx<nri; iidx++)
 472     {
 473         /* Load shift vector for this list */
 474         i_shift_offset   = DIM*shiftidx[iidx];
 475
 476         /* Load limits for loop over neighbors */
 477         j_index_start    = jindex[iidx];
 478         j_index_end      = jindex[iidx+1];
 479
 480         /* Get outer coordinate index */
 481         inr              = iinr[iidx];
 482         i_coord_offset   = DIM*inr;
 483
 484         /* Load i particle coords and add shift vector */
 485         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 486
 487         fix0             = _mm_setzero_pd();
 488         fiy0             = _mm_setzero_pd();
 489         fiz0             = _mm_setzero_pd();
 490
 491         /* Load parameters for i particles */
 492         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 493         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 494
 495         /* Start inner kernel loop */
 496         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrA             = jjnr[jidx];
 501             jnrB             = jjnr[jidx+1];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504
 505             /* load j atom coordinates */
 506             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm_sub_pd(ix0,jx0);
 511             dy00             = _mm_sub_pd(iy0,jy0);
 512             dz00             = _mm_sub_pd(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 516
 517             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 518
 519             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 523             vdwjidx0A        = 2*vdwtype[jnrA+0];
 524             vdwjidx0B        = 2*vdwtype[jnrB+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r00              = _mm_mul_pd(rsq00,rinv00);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq00             = _mm_mul_pd(iq0,jq0);
 534             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 535                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 536
 537             /* Calculate table index by multiplying r with table scale and truncate to integer */
 538             rt               = _mm_mul_pd(r00,vftabscale);
 539             vfitab           = _mm_cvttpd_epi32(rt);
 540 #ifdef __XOP__
 541             vfeps            = _mm_frcz_pd(rt);
 542 #else
 543             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 544 #endif
 545             twovfeps         = _mm_add_pd(vfeps,vfeps);
 546             vfitab           = _mm_slli_epi32(vfitab,3);
 547
 548             /* COULOMB ELECTROSTATICS */
 549             velec            = _mm_mul_pd(qq00,rinv00);
 550             felec            = _mm_mul_pd(velec,rinvsq00);
 551
 552             /* CUBIC SPLINE TABLE DISPERSION */
 553             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 554             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 555             GMX_MM_TRANSPOSE2_PD(Y,F);
 556             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 557             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 558             GMX_MM_TRANSPOSE2_PD(G,H);
 559             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 560             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 561             fvdw6            = _mm_mul_pd(c6_00,FF);
 562
 563             /* CUBIC SPLINE TABLE REPULSION */
 564             vfitab           = _mm_add_epi32(vfitab,ifour);
 565             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 566             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 567             GMX_MM_TRANSPOSE2_PD(Y,F);
 568             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 569             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 570             GMX_MM_TRANSPOSE2_PD(G,H);
 571             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 572             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 573             fvdw12           = _mm_mul_pd(c12_00,FF);
 574             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 575
 576             fscal            = _mm_add_pd(felec,fvdw);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 580             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 581             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 582
 583             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 584                                                    _mm_mul_pd(dx00,fscal),
 585                                                    _mm_mul_pd(dy00,fscal),
 586                                                    _mm_mul_pd(dz00,fscal));
 587
 588             /* Inner loop uses 57 flops */
 589         }
 590
 591         if(jidx<j_index_end)
 592         {
 593
 594             jnrA             = jjnr[jidx];
 595             j_coord_offsetA  = DIM*jnrA;
 596
 597             /* load j atom coordinates */
 598             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 599                                               &jx0,&jy0,&jz0);
 600
 601             /* Calculate displacement vector */
 602             dx00             = _mm_sub_pd(ix0,jx0);
 603             dy00             = _mm_sub_pd(iy0,jy0);
 604             dz00             = _mm_sub_pd(iz0,jz0);
 605
 606             /* Calculate squared distance and things based on it */
 607             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 608
 609             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 610
 611             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 612
 613             /* Load parameters for j particles */
 614             jq0              = _mm_load_sd(charge+jnrA+0);
 615             vdwjidx0A        = 2*vdwtype[jnrA+0];
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r00              = _mm_mul_pd(rsq00,rinv00);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq00             = _mm_mul_pd(iq0,jq0);
 625             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 626
 627             /* Calculate table index by multiplying r with table scale and truncate to integer */
 628             rt               = _mm_mul_pd(r00,vftabscale);
 629             vfitab           = _mm_cvttpd_epi32(rt);
 630 #ifdef __XOP__
 631             vfeps            = _mm_frcz_pd(rt);
 632 #else
 633             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 634 #endif
 635             twovfeps         = _mm_add_pd(vfeps,vfeps);
 636             vfitab           = _mm_slli_epi32(vfitab,3);
 637
 638             /* COULOMB ELECTROSTATICS */
 639             velec            = _mm_mul_pd(qq00,rinv00);
 640             felec            = _mm_mul_pd(velec,rinvsq00);
 641
 642             /* CUBIC SPLINE TABLE DISPERSION */
 643             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 644             F                = _mm_setzero_pd();
 645             GMX_MM_TRANSPOSE2_PD(Y,F);
 646             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 647             H                = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(G,H);
 649             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 650             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 651             fvdw6            = _mm_mul_pd(c6_00,FF);
 652
 653             /* CUBIC SPLINE TABLE REPULSION */
 654             vfitab           = _mm_add_epi32(vfitab,ifour);
 655             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 656             F                = _mm_setzero_pd();
 657             GMX_MM_TRANSPOSE2_PD(Y,F);
 658             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 659             H                = _mm_setzero_pd();
 660             GMX_MM_TRANSPOSE2_PD(G,H);
 661             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 662             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 663             fvdw12           = _mm_mul_pd(c12_00,FF);
 664             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 665
 666             fscal            = _mm_add_pd(felec,fvdw);
 667
 668             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 669
 670             /* Update vectorial force */
 671             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 672             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 673             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 674
 675             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 676                                                    _mm_mul_pd(dx00,fscal),
 677                                                    _mm_mul_pd(dy00,fscal),
 678                                                    _mm_mul_pd(dz00,fscal));
 679
 680             /* Inner loop uses 57 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         /* Increment number of inner iterations */
 689         inneriter                  += j_index_end - j_index_start;
 690
 691         /* Outer loop uses 7 flops */
 692     }
 693
 694     /* Increment number of outer iterations */
 695     outeriter        += nri;
 696
 697     /* Update outer/inner flops */
 698
 699     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 700 }