src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  95     real             *vftab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_pd();
 155         vvdwsum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = avx128fma_invsqrt_d(rsq00);
 180
 181             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _mm_mul_pd(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_pd(iq0,jq0);
 196             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _mm_mul_pd(r00,vftabscale);
 201             vfitab           = _mm_cvttpd_epi32(rt);
 202 #ifdef __XOP__
 203             vfeps            = _mm_frcz_pd(rt);
 204 #else
 205             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 206 #endif
 207             twovfeps         = _mm_add_pd(vfeps,vfeps);
 208             vfitab           = _mm_slli_epi32(vfitab,2);
 209
 210             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 211             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 212             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 213             GMX_MM_TRANSPOSE2_PD(Y,F);
 214             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 215             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 216             GMX_MM_TRANSPOSE2_PD(G,H);
 217             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 218             VV               = _mm_macc_pd(vfeps,Fp,Y);
 219             velec            = _mm_mul_pd(qq00,VV);
 220             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 221             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 227             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 228             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 229             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm_add_pd(velecsum,velec);
 233             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = _mm_add_pd(felec,fvdw);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 239             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 240             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 243                                                    _mm_mul_pd(dx00,fscal),
 244                                                    _mm_mul_pd(dy00,fscal),
 245                                                    _mm_mul_pd(dz00,fscal));
 246
 247             /* Inner loop uses 59 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = avx128fma_invsqrt_d(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = _mm_load_sd(charge+jnrA+0);
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r00              = _mm_mul_pd(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq00             = _mm_mul_pd(iq0,jq0);
 284             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 285
 286             /* Calculate table index by multiplying r with table scale and truncate to integer */
 287             rt               = _mm_mul_pd(r00,vftabscale);
 288             vfitab           = _mm_cvttpd_epi32(rt);
 289 #ifdef __XOP__
 290             vfeps            = _mm_frcz_pd(rt);
 291 #else
 292             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 293 #endif
 294             twovfeps         = _mm_add_pd(vfeps,vfeps);
 295             vfitab           = _mm_slli_epi32(vfitab,2);
 296
 297             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 298             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 299             F                = _mm_setzero_pd();
 300             GMX_MM_TRANSPOSE2_PD(Y,F);
 301             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 302             H                = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(G,H);
 304             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 305             VV               = _mm_macc_pd(vfeps,Fp,Y);
 306             velec            = _mm_mul_pd(qq00,VV);
 307             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 308             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 309
 310             /* LENNARD-JONES DISPERSION/REPULSION */
 311
 312             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 313             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 314             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 315             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 316             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 322             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 323
 324             fscal            = _mm_add_pd(felec,fvdw);
 325
 326             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 330             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 331             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 334                                                    _mm_mul_pd(dx00,fscal),
 335                                                    _mm_mul_pd(dy00,fscal),
 336                                                    _mm_mul_pd(dz00,fscal));
 337
 338             /* Inner loop uses 59 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 349         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 350
 351         /* Increment number of inner iterations */
 352         inneriter                  += j_index_end - j_index_start;
 353
 354         /* Outer loop uses 9 flops */
 355     }
 356
 357     /* Increment number of outer iterations */
 358     outeriter        += nri;
 359
 360     /* Update outer/inner flops */
 361
 362     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 363 }
 364 /*
 365  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 366  * Electrostatics interaction: CubicSplineTable
 367  * VdW interaction:            LennardJones
 368  * Geometry:                   Particle-Particle
 369  * Calculate force/pot:        Force
 370  */
 371 void
 372 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 373                     (t_nblist                    * gmx_restrict       nlist,
 374                      rvec                        * gmx_restrict          xx,
 375                      rvec                        * gmx_restrict          ff,
 376                      struct t_forcerec           * gmx_restrict          fr,
 377                      t_mdatoms                   * gmx_restrict     mdatoms,
 378                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 379                      t_nrnb                      * gmx_restrict        nrnb)
 380 {
 381     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 382      * just 0 for non-waters.
 383      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 384      * jnr indices corresponding to data put in the four positions in the SIMD register.
 385      */
 386     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 387     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 388     int              jnrA,jnrB;
 389     int              j_coord_offsetA,j_coord_offsetB;
 390     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 391     real             rcutoff_scalar;
 392     real             *shiftvec,*fshift,*x,*f;
 393     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     int              vdwioffset0;
 395     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B;
 397     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     int              nvdwtype;
 402     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 403     int              *vdwtype;
 404     real             *vdwparam;
 405     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 406     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 407     __m128i          vfitab;
 408     __m128i          ifour       = _mm_set1_epi32(4);
 409     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 410     real             *vftab;
 411     __m128d          dummy_mask,cutoff_mask;
 412     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 413     __m128d          one     = _mm_set1_pd(1.0);
 414     __m128d          two     = _mm_set1_pd(2.0);
 415     x                = xx[0];
 416     f                = ff[0];
 417
 418     nri              = nlist->nri;
 419     iinr             = nlist->iinr;
 420     jindex           = nlist->jindex;
 421     jjnr             = nlist->jjnr;
 422     shiftidx         = nlist->shift;
 423     gid              = nlist->gid;
 424     shiftvec         = fr->shift_vec[0];
 425     fshift           = fr->fshift[0];
 426     facel            = _mm_set1_pd(fr->ic->epsfac);
 427     charge           = mdatoms->chargeA;
 428     nvdwtype         = fr->ntype;
 429     vdwparam         = fr->nbfp;
 430     vdwtype          = mdatoms->typeA;
 431
 432     vftab            = kernel_data->table_elec->data;
 433     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 434
 435     /* Avoid stupid compiler warnings */
 436     jnrA = jnrB = 0;
 437     j_coord_offsetA = 0;
 438     j_coord_offsetB = 0;
 439
 440     outeriter        = 0;
 441     inneriter        = 0;
 442
 443     /* Start outer loop over neighborlists */
 444     for(iidx=0; iidx<nri; iidx++)
 445     {
 446         /* Load shift vector for this list */
 447         i_shift_offset   = DIM*shiftidx[iidx];
 448
 449         /* Load limits for loop over neighbors */
 450         j_index_start    = jindex[iidx];
 451         j_index_end      = jindex[iidx+1];
 452
 453         /* Get outer coordinate index */
 454         inr              = iinr[iidx];
 455         i_coord_offset   = DIM*inr;
 456
 457         /* Load i particle coords and add shift vector */
 458         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 459
 460         fix0             = _mm_setzero_pd();
 461         fiy0             = _mm_setzero_pd();
 462         fiz0             = _mm_setzero_pd();
 463
 464         /* Load parameters for i particles */
 465         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 466         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 467
 468         /* Start inner kernel loop */
 469         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrA             = jjnr[jidx];
 474             jnrB             = jjnr[jidx+1];
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477
 478             /* load j atom coordinates */
 479             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                               &jx0,&jy0,&jz0);
 481
 482             /* Calculate displacement vector */
 483             dx00             = _mm_sub_pd(ix0,jx0);
 484             dy00             = _mm_sub_pd(iy0,jy0);
 485             dz00             = _mm_sub_pd(iz0,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 489
 490             rinv00           = avx128fma_invsqrt_d(rsq00);
 491
 492             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497             vdwjidx0B        = 2*vdwtype[jnrB+0];
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             r00              = _mm_mul_pd(rsq00,rinv00);
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             qq00             = _mm_mul_pd(iq0,jq0);
 507             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 508                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 509
 510             /* Calculate table index by multiplying r with table scale and truncate to integer */
 511             rt               = _mm_mul_pd(r00,vftabscale);
 512             vfitab           = _mm_cvttpd_epi32(rt);
 513 #ifdef __XOP__
 514             vfeps            = _mm_frcz_pd(rt);
 515 #else
 516             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 517 #endif
 518             twovfeps         = _mm_add_pd(vfeps,vfeps);
 519             vfitab           = _mm_slli_epi32(vfitab,2);
 520
 521             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 522             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 523             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 524             GMX_MM_TRANSPOSE2_PD(Y,F);
 525             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 526             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 527             GMX_MM_TRANSPOSE2_PD(G,H);
 528             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 529             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 530             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 531
 532             /* LENNARD-JONES DISPERSION/REPULSION */
 533
 534             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 535             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 536
 537             fscal            = _mm_add_pd(felec,fvdw);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 543
 544             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 545                                                    _mm_mul_pd(dx00,fscal),
 546                                                    _mm_mul_pd(dy00,fscal),
 547                                                    _mm_mul_pd(dz00,fscal));
 548
 549             /* Inner loop uses 50 flops */
 550         }
 551
 552         if(jidx<j_index_end)
 553         {
 554
 555             jnrA             = jjnr[jidx];
 556             j_coord_offsetA  = DIM*jnrA;
 557
 558             /* load j atom coordinates */
 559             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 560                                               &jx0,&jy0,&jz0);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm_sub_pd(ix0,jx0);
 564             dy00             = _mm_sub_pd(iy0,jy0);
 565             dz00             = _mm_sub_pd(iz0,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 569
 570             rinv00           = avx128fma_invsqrt_d(rsq00);
 571
 572             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 573
 574             /* Load parameters for j particles */
 575             jq0              = _mm_load_sd(charge+jnrA+0);
 576             vdwjidx0A        = 2*vdwtype[jnrA+0];
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r00              = _mm_mul_pd(rsq00,rinv00);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq00             = _mm_mul_pd(iq0,jq0);
 586             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 587
 588             /* Calculate table index by multiplying r with table scale and truncate to integer */
 589             rt               = _mm_mul_pd(r00,vftabscale);
 590             vfitab           = _mm_cvttpd_epi32(rt);
 591 #ifdef __XOP__
 592             vfeps            = _mm_frcz_pd(rt);
 593 #else
 594             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 595 #endif
 596             twovfeps         = _mm_add_pd(vfeps,vfeps);
 597             vfitab           = _mm_slli_epi32(vfitab,2);
 598
 599             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 600             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 601             F                = _mm_setzero_pd();
 602             GMX_MM_TRANSPOSE2_PD(Y,F);
 603             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 604             H                = _mm_setzero_pd();
 605             GMX_MM_TRANSPOSE2_PD(G,H);
 606             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 607             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 608             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 609
 610             /* LENNARD-JONES DISPERSION/REPULSION */
 611
 612             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 613             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 614
 615             fscal            = _mm_add_pd(felec,fvdw);
 616
 617             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 618
 619             /* Update vectorial force */
 620             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 621             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 622             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 623
 624             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 625                                                    _mm_mul_pd(dx00,fscal),
 626                                                    _mm_mul_pd(dy00,fscal),
 627                                                    _mm_mul_pd(dz00,fscal));
 628
 629             /* Inner loop uses 50 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 7 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 649 }