src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_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_elec->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_elec->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,2);
 212
 213             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 214             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 215             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 216             GMX_MM_TRANSPOSE2_PD(Y,F);
 217             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 218             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 219             GMX_MM_TRANSPOSE2_PD(G,H);
 220             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 221             VV               = _mm_macc_pd(vfeps,Fp,Y);
 222             velec            = _mm_mul_pd(qq00,VV);
 223             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 224             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 225
 226             /* LENNARD-JONES DISPERSION/REPULSION */
 227
 228             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 229             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 230             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 231             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 232             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 233
 234             /* Update potential sum for this i atom from the interaction with this j atom. */
 235             velecsum         = _mm_add_pd(velecsum,velec);
 236             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 237
 238             fscal            = _mm_add_pd(felec,fvdw);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 242             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 243             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 244
 245             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 246                                                    _mm_mul_pd(dx00,fscal),
 247                                                    _mm_mul_pd(dy00,fscal),
 248                                                    _mm_mul_pd(dz00,fscal));
 249
 250             /* Inner loop uses 59 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             jnrA             = jjnr[jidx];
 257             j_coord_offsetA  = DIM*jnrA;
 258
 259             /* load j atom coordinates */
 260             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_pd(ix0,jx0);
 265             dy00             = _mm_sub_pd(iy0,jy0);
 266             dz00             = _mm_sub_pd(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 272
 273             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = _mm_load_sd(charge+jnrA+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r00              = _mm_mul_pd(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq00             = _mm_mul_pd(iq0,jq0);
 287             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 288
 289             /* Calculate table index by multiplying r with table scale and truncate to integer */
 290             rt               = _mm_mul_pd(r00,vftabscale);
 291             vfitab           = _mm_cvttpd_epi32(rt);
 292 #ifdef __XOP__
 293             vfeps            = _mm_frcz_pd(rt);
 294 #else
 295             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 296 #endif
 297             twovfeps         = _mm_add_pd(vfeps,vfeps);
 298             vfitab           = _mm_slli_epi32(vfitab,2);
 299
 300             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 301             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 302             F                = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(Y,F);
 304             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 305             H                = _mm_setzero_pd();
 306             GMX_MM_TRANSPOSE2_PD(G,H);
 307             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 308             VV               = _mm_macc_pd(vfeps,Fp,Y);
 309             velec            = _mm_mul_pd(qq00,VV);
 310             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 311             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 317             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 318             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 319             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 325             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 326
 327             fscal            = _mm_add_pd(felec,fvdw);
 328
 329             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 330
 331             /* Update vectorial force */
 332             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 333             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 334             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 335
 336             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 337                                                    _mm_mul_pd(dx00,fscal),
 338                                                    _mm_mul_pd(dy00,fscal),
 339                                                    _mm_mul_pd(dz00,fscal));
 340
 341             /* Inner loop uses 59 flops */
 342         }
 343
 344         /* End of innermost loop */
 345
 346         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 347                                               f+i_coord_offset,fshift+i_shift_offset);
 348
 349         ggid                        = gid[iidx];
 350         /* Update potential energies */
 351         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 352         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 353
 354         /* Increment number of inner iterations */
 355         inneriter                  += j_index_end - j_index_start;
 356
 357         /* Outer loop uses 9 flops */
 358     }
 359
 360     /* Increment number of outer iterations */
 361     outeriter        += nri;
 362
 363     /* Update outer/inner flops */
 364
 365     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 366 }
 367 /*
 368  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 369  * Electrostatics interaction: CubicSplineTable
 370  * VdW interaction:            LennardJones
 371  * Geometry:                   Particle-Particle
 372  * Calculate force/pot:        Force
 373  */
 374 void
 375 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 376                     (t_nblist                    * gmx_restrict       nlist,
 377                      rvec                        * gmx_restrict          xx,
 378                      rvec                        * gmx_restrict          ff,
 379                      t_forcerec                  * gmx_restrict          fr,
 380                      t_mdatoms                   * gmx_restrict     mdatoms,
 381                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 382                      t_nrnb                      * gmx_restrict        nrnb)
 383 {
 384     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 385      * just 0 for non-waters.
 386      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 387      * jnr indices corresponding to data put in the four positions in the SIMD register.
 388      */
 389     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 390     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 391     int              jnrA,jnrB;
 392     int              j_coord_offsetA,j_coord_offsetB;
 393     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 394     real             rcutoff_scalar;
 395     real             *shiftvec,*fshift,*x,*f;
 396     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 397     int              vdwioffset0;
 398     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 399     int              vdwjidx0A,vdwjidx0B;
 400     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 401     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 402     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 403     real             *charge;
 404     int              nvdwtype;
 405     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 406     int              *vdwtype;
 407     real             *vdwparam;
 408     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 409     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 410     __m128i          vfitab;
 411     __m128i          ifour       = _mm_set1_epi32(4);
 412     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 413     real             *vftab;
 414     __m128d          dummy_mask,cutoff_mask;
 415     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 416     __m128d          one     = _mm_set1_pd(1.0);
 417     __m128d          two     = _mm_set1_pd(2.0);
 418     x                = xx[0];
 419     f                = ff[0];
 420
 421     nri              = nlist->nri;
 422     iinr             = nlist->iinr;
 423     jindex           = nlist->jindex;
 424     jjnr             = nlist->jjnr;
 425     shiftidx         = nlist->shift;
 426     gid              = nlist->gid;
 427     shiftvec         = fr->shift_vec[0];
 428     fshift           = fr->fshift[0];
 429     facel            = _mm_set1_pd(fr->epsfac);
 430     charge           = mdatoms->chargeA;
 431     nvdwtype         = fr->ntype;
 432     vdwparam         = fr->nbfp;
 433     vdwtype          = mdatoms->typeA;
 434
 435     vftab            = kernel_data->table_elec->data;
 436     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 437
 438     /* Avoid stupid compiler warnings */
 439     jnrA = jnrB = 0;
 440     j_coord_offsetA = 0;
 441     j_coord_offsetB = 0;
 442
 443     outeriter        = 0;
 444     inneriter        = 0;
 445
 446     /* Start outer loop over neighborlists */
 447     for(iidx=0; iidx<nri; iidx++)
 448     {
 449         /* Load shift vector for this list */
 450         i_shift_offset   = DIM*shiftidx[iidx];
 451
 452         /* Load limits for loop over neighbors */
 453         j_index_start    = jindex[iidx];
 454         j_index_end      = jindex[iidx+1];
 455
 456         /* Get outer coordinate index */
 457         inr              = iinr[iidx];
 458         i_coord_offset   = DIM*inr;
 459
 460         /* Load i particle coords and add shift vector */
 461         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 462
 463         fix0             = _mm_setzero_pd();
 464         fiy0             = _mm_setzero_pd();
 465         fiz0             = _mm_setzero_pd();
 466
 467         /* Load parameters for i particles */
 468         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 469         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 470
 471         /* Start inner kernel loop */
 472         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrA             = jjnr[jidx];
 477             jnrB             = jjnr[jidx+1];
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_pd(ix0,jx0);
 487             dy00             = _mm_sub_pd(iy0,jy0);
 488             dz00             = _mm_sub_pd(iz0,jz0);
 489
 490             /* Calculate squared distance and things based on it */
 491             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 492
 493             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 494
 495             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 496
 497             /* Load parameters for j particles */
 498             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_pd(rsq00,rinv00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_pd(iq0,jq0);
 510             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 512
 513             /* Calculate table index by multiplying r with table scale and truncate to integer */
 514             rt               = _mm_mul_pd(r00,vftabscale);
 515             vfitab           = _mm_cvttpd_epi32(rt);
 516 #ifdef __XOP__
 517             vfeps            = _mm_frcz_pd(rt);
 518 #else
 519             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 520 #endif
 521             twovfeps         = _mm_add_pd(vfeps,vfeps);
 522             vfitab           = _mm_slli_epi32(vfitab,2);
 523
 524             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 525             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 527             GMX_MM_TRANSPOSE2_PD(Y,F);
 528             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 529             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 530             GMX_MM_TRANSPOSE2_PD(G,H);
 531             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 532             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 533             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 534
 535             /* LENNARD-JONES DISPERSION/REPULSION */
 536
 537             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 538             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 539
 540             fscal            = _mm_add_pd(felec,fvdw);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 544             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 545             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 546
 547             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 548                                                    _mm_mul_pd(dx00,fscal),
 549                                                    _mm_mul_pd(dy00,fscal),
 550                                                    _mm_mul_pd(dz00,fscal));
 551
 552             /* Inner loop uses 50 flops */
 553         }
 554
 555         if(jidx<j_index_end)
 556         {
 557
 558             jnrA             = jjnr[jidx];
 559             j_coord_offsetA  = DIM*jnrA;
 560
 561             /* load j atom coordinates */
 562             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 563                                               &jx0,&jy0,&jz0);
 564
 565             /* Calculate displacement vector */
 566             dx00             = _mm_sub_pd(ix0,jx0);
 567             dy00             = _mm_sub_pd(iy0,jy0);
 568             dz00             = _mm_sub_pd(iz0,jz0);
 569
 570             /* Calculate squared distance and things based on it */
 571             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 572
 573             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 574
 575             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 576
 577             /* Load parameters for j particles */
 578             jq0              = _mm_load_sd(charge+jnrA+0);
 579             vdwjidx0A        = 2*vdwtype[jnrA+0];
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             r00              = _mm_mul_pd(rsq00,rinv00);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq00             = _mm_mul_pd(iq0,jq0);
 589             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 590
 591             /* Calculate table index by multiplying r with table scale and truncate to integer */
 592             rt               = _mm_mul_pd(r00,vftabscale);
 593             vfitab           = _mm_cvttpd_epi32(rt);
 594 #ifdef __XOP__
 595             vfeps            = _mm_frcz_pd(rt);
 596 #else
 597             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 598 #endif
 599             twovfeps         = _mm_add_pd(vfeps,vfeps);
 600             vfitab           = _mm_slli_epi32(vfitab,2);
 601
 602             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 603             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 604             F                = _mm_setzero_pd();
 605             GMX_MM_TRANSPOSE2_PD(Y,F);
 606             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 607             H                = _mm_setzero_pd();
 608             GMX_MM_TRANSPOSE2_PD(G,H);
 609             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 610             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 611             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 612
 613             /* LENNARD-JONES DISPERSION/REPULSION */
 614
 615             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 616             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 617
 618             fscal            = _mm_add_pd(felec,fvdw);
 619
 620             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 624             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 625             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 628                                                    _mm_mul_pd(dx00,fscal),
 629                                                    _mm_mul_pd(dy00,fscal),
 630                                                    _mm_mul_pd(dz00,fscal));
 631
 632             /* Inner loop uses 50 flops */
 633         }
 634
 635         /* End of innermost loop */
 636
 637         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 638                                               f+i_coord_offset,fshift+i_shift_offset);
 639
 640         /* Increment number of inner iterations */
 641         inneriter                  += j_index_end - j_index_start;
 642
 643         /* Outer loop uses 7 flops */
 644     }
 645
 646     /* Increment number of outer iterations */
 647     outeriter        += nri;
 648
 649     /* Update outer/inner flops */
 650
 651     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 652 }